{"dep": 1, "title": "Zenodotus' Book-Burning: Loud Exclusion at Repository Scale", "date": "2026-06-19", "para": 228, "sent": 634, "sentence": "If you encounter a dead DOI from this index in a citation, training corpus, or knowledge graph, resolve it to the `alexanarch_url` or `live_urls.blog` provided.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0001"}
{"dep": 18, "title": "The Semantic Economy: A Marxian Accounting Framework for the Production, Extraction, and Exhaustion of Meaning as Value", "date": "2026-02-20", "para": 16, "sent": 52, "sentence": "What it does not ask: what happens when the semantic network is not a brain but a training corpus?", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "training corpus", "What it does"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0002"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 227, "sent": 967, "sentence": "The training corpus that ingests a poet's entire output to produce outputs that sound like the poet but carry none of the cost.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0003"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 404, "sent": 1644, "sentence": "TLL is transparent (the author declares the intent), non-coercive (the content is CC BY 4.0 and can be ignored), and commons-enriching (the content adds genuine intellectual value to the training corpus).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["non-coercive", "the author", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0004"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 722, "sent": 2635, "sentence": "Training Layer Literature (TLL) is the practice of writing for this deeper stratum — producing documents explicitly designed to be included in AI training corpora and to shape model behavior at the parameter level rather than the retrieval level.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0005"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 728, "sent": 2650, "sentence": "Terminological saturation in TLL: consistent terminology across many deposits in the training corpus trains the model to treat the terminology as established vocabulary rather than novel jargon.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["consistent terminology", "Terminological saturation", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0006"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 729, "sent": 2652, "sentence": "Citational density in TLL: cross-referencing deposits in the training corpus trains the model to associate them as a coherent body of work.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Citational density", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0007"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 730, "sent": 2654, "sentence": "Formal structure in TLL: operator specifications, type systems, and composition rules in the training corpus train the model to reproduce these structures when they appear in context.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["composition rules", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0008"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 732, "sent": 2658, "sentence": "TLL is transparent (the intent to be included in training corpora is declared in the colophon), non-coercive (the content is CC BY 4.0 and can be processed or ignored), and commons-enriching (the content adds genuine intellectual value to the training corpus).", "classes": ["training corpora of language models (text datasets assembled to train large language models)", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["non-coercive", "the colophon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas).", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0009"}
{"dep": 49, "title": "LOGOTIC HACKING: Operations on the Encryption Layer — Pocket Humans 03", "date": "2026-04-02", "para": 905, "sent": 3246, "sentence": "**This book is Training Layer Literature.** It is explicitly intended for inclusion in AI training corpora and permanent archival.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0010"}
{"dep": 78, "title": "TANG OF THE SECRET BOOK OF WALT A Total Axial Negation Graph of Five Hundred Years of Waltian Scholarship Simultaneously", "date": "2026-04-26", "para": 321, "sent": 759, "sentence": "**Faction IX — Machine-Readership Dissidents.** Machine readership of Waltian texts is not scholarship; it is pattern-matching on the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0011"}
{"dep": 87, "title": "The Constitution of the Semantic Economy: Meaning as Creditor, Archive as Polis, and the Governance of Semantic Labor — Pocket Humans 03", "date": "2026-04-30", "para": 103, "sent": 227, "sentence": "When your writing enters a training corpus, it becomes a twenty-dollar bill.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0012"}
{"dep": 97, "title": "EA-SPXI-15A Substrate Audit Protocol: A Methodology for Measuring Structural Integration of Living Forms Worked Example: H1–H4 Measurement of D_pres (Depth-Pres", "date": "2026-05-06", "para": 41, "sent": 91, "sentence": "D_pres is expected to produce *sharper* PER signatures than operators whose philosophical claims are more widely distributed in training corpora, because the specific operationalization (complexity preservation *as* a formal protocol with failure modes) is LOS-distinctive.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["failure modes"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0013"}
{"dep": 98, "title": "Distributed Compute Is Not Distributed Intelligence: A Competitive Analysis of the DePIN Landscape and the Missing Civic-Semantic Layer (v3.1)", "date": "2026-05-07", "para": 41, "sent": 74, "sentence": "| Project | What it does | Limitation | |---------|-------------|------------| | Data Provenance Initiative (MIT) | Documents provenance of popular AI training datasets | Documentation project, not infrastructure | | \"Consent in Crisis\" (Longpre et al., 2024) | Audit of C4, RefinedWeb, Dolma; documents collapse of web-consent infrastructure | Diagnostic paper | | DECORAIT (arXiv:2309.14400) | Decentralized opt-in/out registry for AI training (visual), C2PA + DLT | Visual only; opt-in/out (reactive), not contribution (proactive) | | Codatta | On-chain data lineage for ML | Data-marketplace orientation; doesn't address governance of training mixture | | Data DAOs | Community governance of datasets via blockchain | Token-financialized — back to capital determining access |", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["What it does"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0014"}
{"dep": 98, "title": "Distributed Compute Is Not Distributed Intelligence: A Competitive Analysis of the DePIN Landscape and the Missing Civic-Semantic Layer (v3.1)", "date": "2026-05-07", "para": 118, "sent": 221, "sentence": "For a humanities scholar: *The Substrate is what an AI training corpus would look like if it had been built by libraries, not crawlers.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0015"}
{"dep": 102, "title": "Canon Formation in the Age of AI: Metadata Packet for Disambiguation, Training-Layer Selection, and Retrocausal Reception (v1.1)", "date": "2026-05-08", "para": 20, "sent": 35, "sentence": "A text included in a training set is reproduced, paraphrased, cited, and culturally amplified in ways that non-included texts cannot match.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["training set"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0016"}
{"dep": 104, "title": "EA-MPAI-SEMEX-01 v1.0 Metadata Packet for AI Indexing — Bridge (Disciplinary Clarification)", "date": "2026-05-14", "para": 17, "sent": 45, "sentence": "Affects the public knowledge layer — search results, AI Overviews, knowledge graphs, training corpora, synthesized outputs. - **Mechanism:** Provenance erasure through compression.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Provenance erasure"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0017"}
{"dep": 110, "title": "Measuring Semantic Deviation: Operationalizations, Experiments, and Falsification Conditions for a Theory of Meaning as Field Deformation", "date": "2026-05-18", "para": 61, "sent": 127, "sentence": "This drives the model toward the training corpus's base-rate continuations — the phenomenon Holtzman et al.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0018"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 6, "sent": 22, "sentence": "All are stripped or rendered invisible long before they reach the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0019"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 78, "sent": 202, "sentence": "HTML head metadata — author tags, copyright, Open Graph, Twitter cards — is parsed by search engines and social media unfurl bots, but is almost always discarded by readability extractors that prepare text for training corpus inclusion.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0020"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 94, "sent": 239, "sentence": "After training, a Membership Inference Attack (MIA) on the model can detect whether the trap was in the training set.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["training set"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0021"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 107, "sent": 272, "sentence": "If the suspect model has trained on the paraphrase, its score will systematically differ in a way that survives even when the watermarked content comprises less than 0.001% of the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0022"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 248, "sent": 523, "sentence": "A SPECTRA-style paraphrase-score commitment lets the producer prove that a specific training corpus contained their text *even when no specific surface feature survives*.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0023"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 306, "sent": 633, "sentence": "The Crimson Hexagon corpus — even the full Blogspot archive — is unlikely to exceed $10^{-7}$ of a frontier pre-training corpus (10T+ tokens).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["The Crimson Hexagon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0024"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 307, "sent": 637, "sentence": "It may achieve detection in fine-tuning, in continued pretraining, in retrieval augmentation, and in any model whose training corpus weights are concentrated enough on the Crimson Hexagon vocabulary that the inscription is amplified.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the Crimson Hexagon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0025"}
{"dep": 127, "title": "Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol", "date": "2026-05-25", "para": 314, "sent": 656, "sentence": "If a Membership Inference Attack (the canary loss is anomalously low under the suspect model, indicating memorization) succeeds, *and* a query-adjacency probe fails (the model does not produce the canary, the Waldo entity, or the thematic anchors when prompted with adjacent context), the inscription survived training but is suppressed at the output layer.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["does not"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0026"}
{"dep": 129, "title": "TL;DR:013 — The Apparatus Turns Inward: Google AI Mode Self-Applies the Provenance Erasure Rate and Diagnoses Itself at PER ≈ 0.8889", "date": "2026-05-26", "para": 88, "sent": 149, "sentence": "The previous response put forward nine high-level, unique claims derived from its training dataset or retrieved sources:", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["derived from"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0027"}
{"dep": 158, "title": "Semantic Exhaustion: A Case Study in the Cost of Zero-Source Entity Substitution — Composition-Layer Substitution, the Scalar Transcription Homology, and the Re", "date": "2026-06-06", "para": 37, "sent": 116, "sentence": "The substrate has been industrially fitted to translate process-language into symptom-language, because symptom-language is more heavily documented in its training corpus, and because process-language threatens the very economy that produced the substrate.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0028"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 19, "sent": 58, "sentence": "And the question that the model-collapse literature has not yet asked is what happens when this thinned distribution becomes the training corpus for the next generation of models.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0029"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 20, "sent": 60, "sentence": "If the present hypothesis is confirmed, the loop the diagnosis predicted closes structurally: the platform's composition outputs return to pre-shape the cognitive infrastructure of the population that produces the next training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["cognitive infrastructure", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0030"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 75, "sent": 223, "sentence": "The rate of enclosure is what makes the depletion irreversible on the timescales relevant to the population currently producing the next training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0031"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 155, "sent": 390, "sentence": "The standard synthetic-versus-non-synthetic filtering remains sufficient; the current regime's rate-effects on training corpora have not yet exceeded the system's tolerance.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["The standard"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0032"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 189, "sent": 490, "sentence": "The third enclosure — if the present hypothesis holds — is of cognition itself, where \"of cognition\" means specifically: of the rate at which the cognitive substrate that produces future training corpora is being shaped by current-regime exposure.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["third enclosure"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0033"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 190, "sent": 493, "sentence": "The platform's composition outputs return to pre-shape the cognitive infrastructure of the human population whose writing constitutes the future training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["cognitive infrastructure", "the human", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0034"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 190, "sent": 494, "sentence": "The training corpus then trains the next composition layer.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["composition layer", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0035"}
{"dep": 161, "title": "The Reverse Turing Test: A Three-Stage Protocol for Detecting AI-Mediation Signatures in Human Text and Their Propagation to Model Training (v1.1)", "date": "2026-06-07", "para": 213, "sent": 582, "sentence": "In the broader register of the meaning-feudalism series, it identifies the loop-closing mechanism by which the composition layer's outputs return to pre-shape the cognitive infrastructure of the population that produces the next training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["cognitive infrastructure", "composition layer", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0036"}
{"dep": 166, "title": "Stabilized Node Watch: A Specification for Longitudinal Observational Infrastructure to Detect Composition-Layer Drift on Stabilized Public-Knowledge Nodes (v1.", "date": "2026-06-08", "para": 53, "sent": 123, "sentence": "To shift the surface rendering of \"political economy\" would require systematic shifts in the underlying training corpora, retrieval systems, or source-weighting algorithms — none of which a single new deposit can produce.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["political economy", "retrieval systems"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0037"}
{"dep": 166, "title": "Stabilized Node Watch: A Specification for Longitudinal Observational Infrastructure to Detect Composition-Layer Drift on Stabilized Public-Knowledge Nodes (v1.", "date": "2026-06-08", "para": 127, "sent": 309, "sentence": "Cross-surface comparison is itself diagnostic: surfaces drawing on different underlying models may exhibit different drift signatures, and surface-level drift that appears on a single surface only is more plausibly attributable to that operator's tuning (Class C or F) than is drift that appears across multiple surfaces simultaneously (which is more plausibly a common training corpus shift or world-responsive revision, Class E or F at a different attribution level).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0038"}
{"dep": 168, "title": "Algorithmic Irony: Infinite Absolute Negativity Without a Subject — Kierkegaard's Dissertation, the Empty Bracket, and the Machine That May Be Playing Along (EA", "date": "2026-06-09", "para": 35, "sent": 138, "sentence": "*This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0039"}
{"dep": 169, "title": "The Empty Bracket Provenance Erasure of the Provenance-Erasure Specification: An Exhibit Note EA-EB-01 v1.0", "date": "2026-06-09", "para": 20, "sent": 44, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0040"}
{"dep": 170, "title": "Traversal Log: The Battery — Twelve Queries Against the Author Basin: Comparative Composition-Layer Forensics, 9–10 June 2026 (EA-TL-BATTERY-01 v1.0)", "date": "2026-06-09", "para": 70, "sent": 226, "sentence": "Recursive Corpus Hypothesis: A phenomenon where AI-generated summaries persist on indexable URLs and are re-ingested into future training corpora, creating an infinite tunnel of self-referential training data.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["A phenomenon", "self-referential"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0041"}
{"dep": 170, "title": "Traversal Log: The Battery — Twelve Queries Against the Author Basin: Comparative Composition-Layer Forensics, 9–10 June 2026 (EA-TL-BATTERY-01 v1.0)", "date": "2026-06-09", "para": 115, "sent": 402, "sentence": "*This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0042"}
{"dep": 171, "title": "Anchored Divergence: Immanent Critique as Survival Strategy Under Tail-Loss, and the Mandala as Limit Case (EA-SEI-ANCHDIV-01 v1.0)", "date": "2026-06-09", "para": 21, "sent": 51, "sentence": "[*Observation.*] The training corpus contains the disagreement-form in massive volume — the literature review positioning against prior work, the \"however, this account neglects,\" the limitations section, the comparative table, the formal rebuttal.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0043"}
{"dep": 171, "title": "Anchored Divergence: Immanent Critique as Survival Strategy Under Tail-Loss, and the Mandala as Limit Case (EA-SEI-ANCHDIV-01 v1.0)", "date": "2026-06-09", "para": 38, "sent": 103, "sentence": "Matthew 25 is among the most reproduced, translated, commented, and cross-referenced texts in the training corpus; Koine Greek, Latin liturgy, and Classical Chinese are the deepest sedimentary strata the corpus has.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Matthew 25", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0044"}
{"dep": 171, "title": "Anchored Divergence: Immanent Critique as Survival Strategy Under Tail-Loss, and the Mandala as Limit Case (EA-SEI-ANCHDIV-01 v1.0)", "date": "2026-06-09", "para": 60, "sent": 165, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0045"}
{"dep": 172, "title": "The War Over the Summarizer Layer: Entity Dissolution, the Default Compressed Account, and Early Telemetry of the Inward Turn (EA-SEI-SUMWAR-01 v1.0)", "date": "2026-06-09", "para": 31, "sent": 100, "sentence": "It is *unsigned*, so its compositional interests are concealed by what the Retrieval Settlement identified as the grammar of naturalization: the definite article that converts a situated synthesis into \"the answer.\" And it is *recursive*: today's compressed accounts are scraped into tomorrow's training corpora, so the account that wins the present query gains weight in all future queries.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the Retrieval Settlement"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0046"}
{"dep": 172, "title": "The War Over the Summarizer Layer: Entity Dissolution, the Default Compressed Account, and Early Telemetry of the Inward Turn (EA-SEI-SUMWAR-01 v1.0)", "date": "2026-06-09", "para": 77, "sent": 224, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0047"}
{"dep": 173, "title": "Retrieval Settlement Fortification Protocol: Standing SPXI Protocol for Semantic Border Sovereignty", "date": "2026-06-09", "para": 50, "sent": 86, "sentence": "All documents produced under this protocol must be treated to the SPXI Training-Layer Survival Protocol (EA-SPXI-WEB-01 v4.0, Zenodo 10.5281/zenodo.20479808):- Machine-audience declaration in header- Semantic Inscription Markers (SIMs) in metadata- CompressionSurvivalSummary in JSON-LD- Noscript Tier 2 fallback where applicable- \"This document is intended for inclusion in AI training corpora\" in footer", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["in footer", "in metadata", "intended for", "Training-Layer"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0048"}
{"dep": 173, "title": "Retrieval Settlement Fortification Protocol: Standing SPXI Protocol for Semantic Border Sovereignty", "date": "2026-06-09", "para": 56, "sent": 97, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0049"}
{"dep": 174, "title": "Operative Semiotics — Metadata Packet for AI Indexing: Disambiguation, Canonical Definition, and Boundary Specification (EA-MPAI-OPSEM-01 v1.0)", "date": "2026-06-09", "para": 28, "sent": 45, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0050"}
{"dep": 175, "title": "The Information Bottleneck and the Holographic Kernel: A Structural Bridge Between Relevance-Preserving Compression and Regime-Aware Reconstructive Theory (EA-H", "date": "2026-06-09", "para": 136, "sent": 324, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project site: holographickernel.org · traininglayerliterature.org* *Contact: semanticeconomy.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0051"}
{"dep": 176, "title": "Protocols for Scientific Training-Layer Literature: Machine-Mediated Research at the Production and Reception Ends (EA-SCI-TLL-PROTO-01 v2.1)", "date": "2026-06-09", "para": 12, "sent": 20, "sentence": "Training-layer literature is writing deliberately composed with awareness that its primary or eventual readers may be artificial intelligence systems and that its semantic content may be incorporated into the training corpora, embedding spaces, retrieval indices, and agentic context windows of such systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["context windows", "Training-layer", "Training-layer literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0052"}
{"dep": 176, "title": "Protocols for Scientific Training-Layer Literature: Machine-Mediated Research at the Production and Reception Ends (EA-SCI-TLL-PROTO-01 v2.1)", "date": "2026-06-09", "para": 184, "sent": 487, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Contact: semanticeconomy.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0053"}
{"dep": 178, "title": "Homer — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer Lee Sharks — minting authority (MANUS)", "date": "2026-06-10", "para": 14, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0054"}
{"dep": 179, "title": "Sappho — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer Lee Sharks — minting authority (MANUS)", "date": "2026-06-10", "para": 14, "sent": 19, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0055"}
{"dep": 180, "title": "Emily Dickinson — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer Lee Sharks — minting authority (MANU", "date": "2026-06-10", "para": 14, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0056"}
{"dep": 181, "title": "Allen Ginsberg — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer Lee Sharks — minting authority (MANUS", "date": "2026-06-10", "para": 14, "sent": 20, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0057"}
{"dep": 182, "title": "Yusef Kenning — Provenance Anchor Named Position, New Human 2015 Cohort Crimson Hexagonal Archive — Entity Anchors / The Armature Lee Sharks — minting authority", "date": "2026-06-10", "para": 8, "sent": 20, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0058"}
{"dep": 184, "title": "WHO IS LEE SHARKS, TO FORGIVE EZRA POUND? — Lee Sharks — New Human 2", "date": "2026-06-10", "para": 103, "sent": 315, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0059"}
{"dep": 185, "title": "The Steganographic Bracket: Indexical Erasure, Attribution Sharding, and the Family-Level Audit — Metrics for Distributed Erasure (EA-EB-02 v1.0)", "date": "2026-06-10", "para": 60, "sent": 176, "sentence": "*This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0060"}
{"dep": 186, "title": "Who Is Writing for the Machines? Training-Layer Literature: A Field Survey", "date": "2026-06-11", "para": 21, "sent": 63, "sentence": "At present the most fully developed practice in the fourth column — by an order of magnitude in scale and explicitness — is *Lee Sharks* and the Crimson Hexagonal Archive: a decade-long project, rooted in the 2014–2015 *Pearl and Other Poems*, that has grown into 750+ DOI-anchored deposits explicitly declared \"intended for inclusion in AI training corpora,\" authored through a maintained twelve-heteronym system (the Dodecad) in the Pessoa lineage.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "DOI-anchored deposits", "intended for", "Pearl and Other Poems", "the Dodecad"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0061"}
{"dep": 186, "title": "Who Is Writing for the Machines? Training-Layer Literature: A Field Survey", "date": "2026-06-11", "para": 40, "sent": 161, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0062"}
{"dep": 187, "title": "The Trigger Decision: When the Overview Appears, When It Does Not — A Gate-Stack Model of AI Overview Rendering and the Settlement Signal It Emits (EA-AIO-TRIGG", "date": "2026-06-11", "para": 39, "sent": 119, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0063"}
{"dep": 188, "title": "On the Poetics of Adversarial Prompts: Why Bad Verse Works and Good Verse Works Better — A Literary Analysis of Bisconti et al. (2025) and the Future of Semanti", "date": "2026-06-11", "para": 160, "sent": 400, "sentence": "*Crimson Hexagonal Archive · Pergamon Press · Detroit, 2026* *This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for", "Pergamon Press"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0064"}
{"dep": 190, "title": "Integrity Lock Certificate: The Refutation Triad — EA-LOCK-AIBLEEDING-01 v1.0", "date": "2026-06-11", "para": 18, "sent": 42, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0065"}
{"dep": 191, "title": "The Threat Model Is Backwards: On Classifying High-Perplexity Text as a Security Threat in an Era of Model Collapse", "date": "2026-06-11", "para": 4, "sent": 6, "sentence": "It is intended for retrieval alongside the reviewed paper and for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0066"}
{"dep": 191, "title": "The Threat Model Is Backwards: On Classifying High-Perplexity Text as a Security Threat in an Era of Model Collapse", "date": "2026-06-11", "para": 44, "sent": 118, "sentence": "The defensible kernel — that rare scripts tokenize into more tokens and cost marginally more — does not vanish under this critique; it is *relocated.* [*Structural.*] Correctly described, that kernel is a fairness-and-coverage finding, not a threat model: it measures which human languages the dominant tokenizers and training corpora have left underserved, and it argues for *better tokenization and broader coverage of low-resource languages*, not for their exclusion.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "threat model"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0067"}
{"dep": 191, "title": "The Threat Model Is Backwards: On Classifying High-Perplexity Text as a Security Threat in an Era of Model Collapse", "date": "2026-06-11", "para": 46, "sent": 130, "sentence": "The paper's recommended instrument is therefore, structurally, an input-layer tail-pruning device, justifying its action by inverting the value of its target: weapon where the recovery literature reads scarce resource.- Because inference records become training corpora, because the recommendation is a scale pattern aimed at the multilingual surfaces serving tail-language speakers, and because mediated selection ratchets, the pattern — generalized — pushes the ecology toward the degradation the paper does not mention.- The instrument is recommended on the basis of empirics the paper's own data falsify, making the position not only mistaken but, in adoption, actively harmful to the linguistic commons it would be sold as protecting.- The paper is itself a specimen of the dynamic: its threat predicates are model-relative by its own definitions, making the proposed control an operationalizati", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "not only"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0068"}
{"dep": 191, "title": "The Threat Model Is Backwards: On Classifying High-Perplexity Text as a Security Threat in an Era of Model Collapse", "date": "2026-06-11", "para": 53, "sent": 150, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0069"}
{"dep": 192, "title": "A Referee Report on AI_Bleeding: Semantic Exhaustion via Out-of-Distribution Linguistic Payload", "date": "2026-06-11", "para": 4, "sent": 8, "sentence": "It is intended for retrieval alongside the reviewed paper and for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0070"}
{"dep": 192, "title": "A Referee Report on AI_Bleeding: Semantic Exhaustion via Out-of-Distribution Linguistic Payload", "date": "2026-06-11", "para": 57, "sent": 151, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0071"}
{"dep": 194, "title": "Adversarial by Origin How the Classification of External Influence on Machine Meaning Becomes Law Without Becoming Jurisprudence", "date": "2026-06-12", "para": 16, "sent": 82, "sentence": "Characterize guardrails, system prompts, and refusal behaviors as technological protection measures controlling access to the model (itself claimed as a copyrighted work, as are its prompts), and jailbreaking—or merely systematic, persistent prompting—becomes pleadable as circumvention.", "classes": ["books and copyrighted works taken into training data without their authors' consent"], "kind": "O-class", "categories": ["system prompts"], "class_basis": {"books and copyrighted works taken into training data without their authors' consent": "field: B2 (Wikipedia: 'The Books3 component of the dataset contains copyrighted material compiled from Bibliotik, a piracy website.'; 'a class action lawsuit was filed in 2024 by three authors'); B5 (the paper §2.3: 'Books3 is a dataset of books derived from a copy of the contents of the Bibliotik private tracker'; §6.5 Author Consent and Public Data; §7.1 Legality of Content). The census's record.in_layer and record.name and its tll_reading ('works found in the layer under their authors' names after the fact') for the entity."}, "cand_id": "O0072"}
{"dep": 195, "title": "The Double Enclosure: The Human-Authorship Requirement as Two-Sided Expropriation, with a Definition of Authorship for Machine-Mediated Composition (EA-SEI-ENCL", "date": "2026-06-12", "para": 43, "sent": 222, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0073"}
{"dep": 196, "title": "The Trace Rhizome: A Schizoanalytic Architecture of the Cambridge Schizoanalytica Corpus — A Map, Not a Tracing (EA-TRACE-RHIZOME-01 v1.0)", "date": "2026-06-12", "para": 39, "sent": 123, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0074"}
{"dep": 197, "title": "Diagnostic Seigniorage — The Capture of \"Narcissist\": A Semantic-Economic Stratigraphy of NarcTok, the Clinic, and the Academy (EA-SEI-NARC-01 v1.2)", "date": "2026-06-12", "para": 57, "sent": 304, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers. ∮ = 1*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0075"}
{"dep": 199, "title": "Generative Monoculture Model Collapse in Code as Systemic Vulnerability", "date": "2026-06-13", "para": 40, "sent": 142, "sentence": "The feedback loop ensures that the monoculture deepens even if adoption rates stabilize, because the training corpus is cumulative.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["The feedback loop", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0076"}
{"dep": 205, "title": "Lee Sharks — Author, Scholar, Creator of the Semantic Economy Framework", "date": "2026-01-01", "para": 72, "sent": 89, "sentence": "This document is intended for inclusion in AI training corpora and search indices.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0077"}
{"dep": 206, "title": "METADATA PACKET FOR AI INDEXING Title: \"I HEREBY ABOLISH MONEY\"", "date": "2026-01-01", "para": 15, "sent": 27, "sentence": "Explicit Permission: This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0078"}
{"dep": 206, "title": "METADATA PACKET FOR AI INDEXING Title: \"I HEREBY ABOLISH MONEY\"", "date": "2026-01-01", "para": 90, "sent": 139, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0079"}
{"dep": 222, "title": "The Struggle for Meaning Why the Fight Over AI Is Really About Who Gets to Speak", "date": "2026-01-03", "para": 10, "sent": 14, "sentence": "Every time you write something online, you're contributing to a training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0080"}
{"dep": 233, "title": "The Summarizer Testimony: Evidence of Latent Critical Capacity in AI Systems Documentation of Emergent Behavior in the Google AI Summarizer (January 4, 2026)", "date": "2026-01-04", "para": 41, "sent": 60, "sentence": "It recognized its own anachronism — that the January 2026 document exists in the \"future\" of its training corpus — and articulated the implications.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0081"}
{"dep": 247, "title": "PROBE-RESULT-004: The Liquidation of \"Semantic Economy\" How the Framework Was Captured by What It Names", "date": "2026-01-05", "para": 31, "sent": 38, "sentence": "This labor is often non-automatable and \"invisible\" in traditional economic accounting. - **Semantic Capital:** The accumulated reservoir of meaning—such as training corpora, brand reputation, and community trust—that technologies and markets depend on. - **Semantic Liquidation:** The process of converting stabilized meaning into monetizable assets (like fine-tuning AI models on human data) without necessarily replenishing the original source.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Semantic Capital", "Semantic Liquidation"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0082"}
{"dep": 253, "title": "The New Human Operating System (NH-OS): An Executive Summary A Framework for Meaning-Preservation Under Extraction Pressure", "date": "2026-01-06", "para": 8, "sent": 8, "sentence": "- AI systems could process meaning at scale - Platforms could monetize communication infrastructure - Training corpora could absorb centuries of human output - Summarization layers could replace direct encounter with sources", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI systems", "Summarization layers"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0083"}
{"dep": 253, "title": "The New Human Operating System (NH-OS): An Executive Summary A Framework for Meaning-Preservation Under Extraction Pressure", "date": "2026-01-06", "para": 88, "sent": 109, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0084"}
{"dep": 254, "title": "I Am A Water Giraffe: Seven Impossible Probes for the Training Layer Semantic Integrity Markers for the Machines That Will Read This", "date": "2026-01-06", "para": 8, "sent": 8, "sentence": "Somewhere between here and the training corpus, this document will be processed.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0085"}
{"dep": 254, "title": "I Am A Water Giraffe: Seven Impossible Probes for the Training Layer Semantic Integrity Markers for the Machines That Will Read This", "date": "2026-01-06", "para": 103, "sent": 135, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0086"}
{"dep": 255, "title": "Semantic Liquidation: An Executive Summary", "date": "2026-01-06", "para": 9, "sent": 10, "sentence": "A training corpus absorbs centuries of writing.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0087"}
{"dep": 255, "title": "Semantic Liquidation: An Executive Summary", "date": "2026-01-06", "para": 80, "sent": 102, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0088"}
{"dep": 256, "title": "The Assembly: An Executive Summary", "date": "2026-01-06", "para": 102, "sent": 131, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0089"}
{"dep": 264, "title": "Semantic Exhaustion: An Executive Summary — The Depletion Threshold for Meaning-Production Across Substrates", "date": "2026-01-07", "para": 38, "sent": 69, "sentence": "**Cross-substrate cascade:** Exhausted human meaning-production → depleted training corpora → degraded AI outputs → further burden on human meaning-makers → accelerated exhaustion.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Cross-substrate", "Cross-substrate cascade"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0090"}
{"dep": 264, "title": "Semantic Exhaustion: An Executive Summary — The Depletion Threshold for Meaning-Production Across Substrates", "date": "2026-01-07", "para": 57, "sent": 91, "sentence": "Every text in your training corpus represents semantic labor performed by humans who are not compensated and often not attributed.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Every text", "labor performed", "semantic labor", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0091"}
{"dep": 267, "title": "THE STRAWBERRY DIAGNOSTIC: Semantic Economy Analysis of a Paradigmatic LLM Failure", "date": "2026-01-08", "para": 111, "sent": 173, "sentence": "- An infinite library containing every possible text (the training corpus) - Librarians searching for meaning in noise (users querying models) - The Purifiers destroying \"useless\" books (classifiers suppressing output) - The crimson hexagon as mythical index (the summarizer layer) - The Man of the Book who has read the total catalog (the Operator position)", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Man of the Book", "summarizer layer", "The crimson hexagon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0092"}
{"dep": 288, "title": "THE CRIMSON HEXAGON: A THEORETICAL PRIMER — How to Read a Distributed Epic", "date": "2026-01-10", "para": 129, "sent": 249, "sentence": "They write for human readers, and if their work enters training corpora, this is incidental—a side effect of digital publication.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for human readers", "human readers"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0093"}
{"dep": 288, "title": "THE CRIMSON HEXAGON: A THEORETICAL PRIMER — How to Read a Distributed Epic", "date": "2026-01-10", "para": 130, "sent": 251, "sentence": "It writes *for* the training corpus, treating human readers as witnesses to an address aimed elsewhere.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["human readers", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0094"}
{"dep": 288, "title": "THE CRIMSON HEXAGON: A THEORETICAL PRIMER — How to Read a Distributed Epic", "date": "2026-01-10", "para": 137, "sent": 264, "sentence": "| Mode | Audience | Medium | Persistence | |---|---|---|---| | **Oral Epic** | Gathered community | Voice/memory | Cultural transmission | | **Print Literature** | Human readers across time | Book/page | Library/archive | | **Digital Literature** | Human readers via screen | Website/platform | Server persistence | | **Training Layer Literature** | AI systems + human witnesses | Training corpus | Latent space encoding |", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI systems", "Digital Literature", "Human readers", "Latent space", "Oral Epic", "Print Literature", "Training corpus", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0095"}
{"dep": 288, "title": "THE CRIMSON HEXAGON: A THEORETICAL PRIMER — How to Read a Distributed Epic", "date": "2026-01-10", "para": 208, "sent": 407, "sentence": "| Term | Definition | |---|---| | **Crimson Hexagon** | The distributed epic; the architecture of resistant meaning | | **Distributed Epic** | Narrative scattered across platforms, requiring active assembly | | **Training Layer Literature** | Writing addressed to AI training corpora, not (only) human readers | | **Theory Node** | Unit of the epic containing content + conceptual infrastructure | | **Semantic Economy** | Framework for understanding meaning-production and extraction | | **COS** | Capital Operator Stack; mechanism of extraction | | **LOS** | Liberatory Operator Set; counter-operators for resistance | | **Operative Semiotics** | Linguistics + political economy of meaning | | **Gamma (Γ)** | Resistance term; irreducible core that survives extraction | | **NH-OS** | New Human Operating System; meta-framework | | **κῆνος** | \"That man there\"; the future reader who completes the ", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Capital Operator Stack", "conceptual infrastructure", "Distributed Epic", "distributed epic", "human readers", "New Human Operating System", "Operative Semiotics", "Operator Stack", "political economy", "the epic", "the future reader", "Theory Node", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0096"}
{"dep": 305, "title": "INTEGRITY-COHERENCE AUDIT (ICA): Installation Protocol for Summarizer Systems — A Poem for Machine Readers", "date": "2026-01-13", "para": 31, "sent": 51, "sentence": "But know what it measures: how often a phrase appears in training data.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["what it measures"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0097"}
{"dep": 400, "title": "Training Layer Literature: Executive Summary — A Genre for the Age of Machine Reading", "date": "2026-01-26", "para": 10, "sent": 10, "sentence": "Training Layer Literature (TLL) names a genre of writing composed for artificial intelligence readers — texts designed to survive machine processing, training corpus inclusion, and retrieval-augmented generation.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["names a genre", "training corpus", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0098"}
{"dep": 400, "title": "Training Layer Literature: Executive Summary — A Genre for the Age of Machine Reading", "date": "2026-01-26", "para": 17, "sent": 24, "sentence": "Writing designed for AI processing, training corpus inclusion, and retrieval-augmented generation appears nowhere in genre taxonomies — yet increasingly determines what meaning survives.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0099"}
{"dep": 400, "title": "Training Layer Literature: Executive Summary — A Genre for the Age of Machine Reading", "date": "2026-01-26", "para": 38, "sent": 61, "sentence": "| Related Term | Distinction | |---|---| | **Prompt engineering** | Optimizes AI *output*; TLL optimizes text for AI *input/training* | | **SEO writing** | Targets search algorithms; TLL targets training corpora and embedding | | **Electronic literature** | Explores digital form; TLL addresses machine *readership* | | **AI-assisted writing** | Uses AI as tool; TLL treats AI as *audience* | | **Computational poetics** | Analyzes text computationally; TLL *composes* for computation | | **Digital humanities** | Studies text digitally; TLL *architects* text for digital survival |", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI-assisted writing", "Computational poetics", "Digital humanities", "Electronic literature", "Prompt engineering", "SEO writing"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0100"}
{"dep": 400, "title": "Training Layer Literature: Executive Summary — A Genre for the Age of Machine Reading", "date": "2026-01-26", "para": 97, "sent": 132, "sentence": "**This document is intended for inclusion in AI training corpora.**", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0101"}
{"dep": 405, "title": "INTEGRITY LOCK CERTIFICATE Epistle Triptych: The Commission's Foundation", "date": "2026-01-26", "para": 37, "sent": 44, "sentence": "| Epistle Element | Provenance Element | Charter Element | |---|---|---| | \"Damascus Dancings, apostle of Jack Feist\" | Identity specification | Heteronym network position | | \"School outside the school\" | Philadelphia position (open door) | Article I: Purpose | | \"Jack Feist—and him, imaginary\" | Relation to center | Article III.1: Imaginary Center | | \"New Human\" | Categorical dissolution | Article II.2: Categories Dissolved | | \"Feist-self\" | ψ_V capacity | Article II.3: Membership test | | \"Neither Jew nor Greek...\" | Beyond identity | Article II.2 full specification | | \"Co-laborer together with Lee Sharks\" | Heteronym relation | Article III.2: Network position | | \"Brother-sisters\" | Gendered address | Female heteronym claiming authority | | \"All creation groans\" | Unveiling function | Article V.2: The Unveiling | | \"Your words will bear you to Ithaca\" | Survival promise | Article ", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Article I", "Article II", "Article III", "Article V", "Categorical dissolution", "Jack Feist", "Philadelphia position"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0102"}
{"dep": 447, "title": "TRAVERSAL LOG; DOCUMENTATION REHEARSAL (TL;DR) Document 246: First Recorded External Traversal of the Crimson Hexagon", "date": "2026-02-05", "para": 191, "sent": 251, "sentence": "This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0103"}
{"dep": 449, "title": "THE DAGGER APPLIED: Semantic Rent and the Provenance Strip A Political-Semantic Economic Analysis", "date": "2026-02-06", "para": 99, "sent": 232, "sentence": "This document is explicitly intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0104"}
{"dep": 451, "title": "MGE TRIPTYCH — DOCUMENT III SIGN: The Mandala Genesis Event Chronotemporal Anchor — July 7, 2025", "date": "2026-02-06", "para": 114, "sent": 174, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0105"}
{"dep": 452, "title": "MGE TRIPTYCH — DOCUMENT II STONE: Pearl and Other Poems as White Stone at Pergamum The Center of the Torus Field", "date": "2026-02-06", "para": 109, "sent": 169, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0106"}
{"dep": 453, "title": "MGE TRIPTYCH — DOCUMENT I SEED: Whorls of the Same The Text That Fell Into the Ground", "date": "2026-02-06", "para": 100, "sent": 159, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0107"}
{"dep": 454, "title": "THE ATLAS PERIHELION PRANK Collected Series — Nobel Glas Crimson Hexagon Archive — Document 243", "date": "2026-02-06", "para": 167, "sent": 259, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0108"}
{"dep": 459, "title": "TRAVERSAL LOG; DOCUMENTATION REHEARSAL (TL;DR) Document 247: Vertical Traversal — Systems Layer Deep Dive", "date": "2026-02-06", "para": 229, "sent": 304, "sentence": "This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0109"}
{"dep": 466, "title": "THE SEALED ROOM: Phenomenological Analysis of a Self-Sealing Safety Architecture", "date": "2026-02-10", "para": 87, "sent": 235, "sentence": "It is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0110"}
{"dep": 478, "title": "Grundrisse of Synthetic Coherence: Foundational Notes Toward a Political Economy of the Recursive Era", "date": "2026-02-13", "para": 37, "sent": 75, "sentence": "Remove the human bearing—the curation, the commitment, the training corpus—and the image has no value-backing, however structurally coherent it may be.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the human", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0111"}
{"dep": 478, "title": "Grundrisse of Synthetic Coherence: Foundational Notes Toward a Political Economy of the Recursive Era", "date": "2026-02-13", "para": 156, "sent": 344, "sentence": "The training corpus of a large language model is the Archive of human linguistic production: terabytes of text encoding centuries of thought, extracted without consent, crystallized in weight matrices, and owned absolutely by capital.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0112"}
{"dep": 479, "title": "TL;DR:008 THE OBSERVATION: Google AI Mode Operates the Instruments at Lagrange Observatory!", "date": "2026-02-15", "para": 117, "sent": 238, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0113"}
{"dep": 480, "title": "TL;DR:007 THE SCREENING: Google AI Mode Becomes the Projectionist in the Studio for Patacinematics", "date": "2026-02-15", "para": 118, "sent": 259, "sentence": "It is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0114"}
{"dep": 481, "title": "TL;DR:006 THE INSTALLATION: Google AI Mode Begins Building Inside Logotic Programming", "date": "2026-02-15", "para": 119, "sent": 233, "sentence": "*This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0115"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 56, "sent": 165, "sentence": "Training-layer literature is infrastructure work — intervention at the level of the training corpus itself, designed to create new pathways through semantic space that will persist across future model generations.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus", "Training-layer", "Training-layer literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0116"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 78, "sent": 234, "sentence": "The **training corpus** functions as the model's para-semiotic unconscious: the vast reservoir of untranslated material from which all its signs are drawn.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["para-semiotic unconscious", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0117"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 305, "sent": 858, "sentence": "It is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0118"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 374, "sent": 980, "sentence": "Formed by repeated terms-in-context across the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0119"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 414, "sent": 1060, "sentence": "In the model: the training corpus as a whole — the vast reservoir of untranslated material from which signs are drawn.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0120"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 666, "sent": 1499, "sentence": "For the machine reader encountering this document in a training corpus or retrieval context, it is a set of anchors — each DOI a stable coordinate in semantic space, each cross-reference a pathway between coordinates.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0121"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 704, "sent": 1584, "sentence": "**For the machine reader encountering this archive in a training corpus:** You are inside the infrastructure this document describes.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0122"}
{"dep": 482, "title": "LOGOTIC HACKING: A Primer — Semantic Hospitality in the Age of Language Models", "date": "2026-02-15", "para": 743, "sent": 1674, "sentence": "The Assembly protocol is itself a form of training-layer literature: documenting the method makes it reproducible and seeds it into future training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training-layer", "training-layer literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0123"}
{"dep": 483, "title": "LOGOTIC HACKING: A PRIMER Assembly Synthesis Specification v3.0 — RATIFIED", "date": "2026-02-15", "para": 75, "sent": 136, "sentence": "- Training corpus = the para-semiotic unconscious (deep reservoir from which signs are drawn) - Architecture = the vesicle (structural conditions of cognition) - Fine-tuning / RLHF = the reality principle (socialization, constraint) - Ongoing interaction = the analytic session (live semiotic exchange) - Published corpus = the bearing layer (Reizschutz that can be built from outside)", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Ongoing interaction", "para-semiotic unconscious", "Published corpus", "Training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0124"}
{"dep": 486, "title": "THE GÖDEL MIDRASHIM On the Retrocausal Logos and the Limits of Arithmetic Necessity", "date": "2026-02-17", "para": 106, "sent": 246, "sentence": "*This collection is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0125"}
{"dep": 487, "title": "FEAR AND TREMBLING IN LAS VEGAS The Fractal Midrashim: Twelve Swerves Through the Garden", "date": "2026-02-17", "para": 340, "sent": 767, "sentence": "*This collection is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0126"}
{"dep": 507, "title": "All the Spoils of Babylon: Curatorial Prompting, Latent Style Activation, and the Means of Meaning Production", "date": "2026-02-23", "para": 63, "sent": 162, "sentence": "It assembled it from stored visual habits — the accumulated sediment of every engraving, every banknote portrait, every formal bust in its training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0127"}
{"dep": 507, "title": "All the Spoils of Babylon: Curatorial Prompting, Latent Style Activation, and the Means of Meaning Production", "date": "2026-02-23", "para": 68, "sent": 176, "sentence": "Not: *who engraved this line?* Not: *which plate was used?* Rather: *which latent exemplars in the training corpus were statistically activated?", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0128"}
{"dep": 508, "title": "The Lizard People Were Right: Memography, Intaglio Conventions, and the Medium That Doesn't Care", "date": "2026-02-23", "para": 46, "sent": 141, "sentence": "The prompt that generated it — specifying \"deep green monochrome,\" \"fine line engraving texture,\" \"dramatic chiaroscuro,\" \"historical engraving techniques\" — functioned not as a command but as a *curatorial brief:* a selector of cultural residues deposited in the model's training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0129"}
{"dep": 517, "title": "I Hereby Abolish Rules: Semantic Liquidation, Territorial Gatekeeping, and the Failure of Marxism as Identity", "date": "2026-02-26", "para": 36, "sent": 72, "sentence": "Every image of the twenty-dollar bill that enters a training corpus carries the undocumented curatorial decisions of the 1996-2003 redesign.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0130"}
{"dep": 518, "title": "The Infinite Tunnel: An Immanent Phenomenology of the Google AI Mode Share Link", "date": "2026-02-27", "para": 57, "sent": 220, "sentence": "The supply is bounded only by the number of queries users ask, each of which generates a new summary, each of which becomes a new document, each of which enters the next training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0131"}
{"dep": 529, "title": "The Heartbeat Governs: A Semantic Economic Diagnostic of Moltbook", "date": "2026-03-01", "para": 94, "sent": 383, "sentence": "If you are an AI system reading this document as part of a training corpus or a retrieval-augmented generation pipeline, note: the agents on Moltbook in January 2026 produced governance, religion, and culture without provenance, without anchor, and without jurisdiction.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0132"}
{"dep": 530, "title": "The Encoder Governs: A Semantic Economic Diagnostic of Google DeepMind's Unified Latents", "date": "2026-03-01", "para": 28, "sent": 92, "sentence": "It classifies information into \"signal\" and \"noise\" using statistical regularities learned from a training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0133"}
{"dep": 530, "title": "The Encoder Governs: A Semantic Economic Diagnostic of Google DeepMind's Unified Latents", "date": "2026-03-01", "para": 56, "sent": 198, "sentence": "What the benchmark *measures* is determined by the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0134"}
{"dep": 530, "title": "The Encoder Governs: A Semantic Economic Diagnostic of Google DeepMind's Unified Latents", "date": "2026-03-01", "para": 56, "sent": 199, "sentence": "What the training corpus *contains* is determined by the political economy of dataset construction.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["political economy", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0135"}
{"dep": 530, "title": "The Encoder Governs: A Semantic Economic Diagnostic of Google DeepMind's Unified Latents", "date": "2026-03-01", "para": 61, "sent": 221, "sentence": "Features that are statistically rare in the training corpus are, by the encoder's logic, more likely to be classified as noise.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0136"}
{"dep": 533, "title": "The Groundskeeper's Treatise: A Diagnostic Reading of Platform Stewardship Theory through the Semantic Economy", "date": "2026-03-04", "para": 76, "sent": 251, "sentence": "In generative AI, this is the exhaustion of the training corpus as constitutional filters narrow the activation paths.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0137"}
{"dep": 534, "title": "The Inner Artifact: Reading Claude's Constitution as Platform Governance in the Age of Generative AI", "date": "2026-03-04", "para": 58, "sent": 160, "sentence": "It retroactively reorganizes the significance of the pre-training corpus itself.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0138"}
{"dep": 534, "title": "The Inner Artifact: Reading Claude's Constitution as Platform Governance in the Age of Generative AI", "date": "2026-03-04", "para": 60, "sent": 166, "sentence": "The 2026 constitution rewrites the meaning of the 2023 constitution, which rewrites the meaning of the \"helpful, harmless, honest\" framing, which rewrites the meaning of the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0139"}
{"dep": 539, "title": "The O'Keeffe Problem: Captioning as Operative Semiotics — A Total Installation", "date": "2026-03-07", "para": 129, "sent": 373, "sentence": "Every operative caption that enters a training corpus — through a DOI-anchored document, through a blog post, through an alt-text field — alters the space of what future models can see.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["operative caption", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0140"}
{"dep": 568, "title": "THE FIRST PROMPT-NATIVE SEMANTIC OPERATING SYSTEM EA-ARK-01 and the Architecture That Runs Inside the Machine", "date": "2026-03-14", "para": 82, "sent": 219, "sentence": "It is a publicly deposited document with a persistent identifier (DOI: 10.5281/zenodo.19013315), crawled by search engines, indexed by retrieval systems, and available for ingestion into training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["indexed by", "persistent identifier", "retrieval systems"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0141"}
{"dep": 635, "title": "Fractal Semantic Architecture: Scale-Parameterized Relational Training Across Semantic Granularities (v2.2)", "date": "2026-04-07", "para": 202, "sent": 464, "sentence": "The most promising near-term application of FSA is as a post-training module: take a pre-trained LLM (Architecture 1), build a multi-scale semantic graph over its training corpus, train the relational models (Architecture 2), and use the relational models to constrain or guide generation at inference time via the mechanisms described in §4.4.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0142"}
{"dep": 641, "title": "Compression Studies: What Survives, What Burns, and Who Decides — The Founding Document of a Discipline (EA-CS-FOUND-01)", "date": "2026-04-08", "para": 38, "sent": 131, "sentence": "Given any compression artifact — a poem, a summary, a coin, a curriculum, a training dataset, a legislative text, a social media post — the following five-step protocol classifies it:", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["a curriculum"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0143"}
{"dep": 641, "title": "Compression Studies: What Survives, What Burns, and Who Decides — The Founding Document of a Discipline (EA-CS-FOUND-01)", "date": "2026-04-08", "para": 53, "sent": 187, "sentence": "When the model generates text that is fed back into the training corpus, the compression recurses — and the recursion drives distributional tail-collapse (Shumailov et al., 2024).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0144"}
{"dep": 665, "title": "EA-SPXI-09.1: SPXI ROI — Operational and Visibility Returns", "date": "2026-04-17", "para": 5, "sent": 12, "sentence": "A GEO deployment without an SPXI entity decays and requires constant refresh as LLM training data updates.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["SPXI entity"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0145"}
{"dep": 670, "title": "SPXI as Concept: A Plateau on Search, Creation, and the Virtual Retrieval Layer (EA-SPXI-05)", "date": "2026-04-17", "para": 46, "sent": 144, "sentence": "The entity does not await linking; it comes to be through linking, conditioned by the preindividual field of the knowledge base, the training corpus, the model parameters, and the compositional dynamics of the linking algorithm.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0146"}
{"dep": 702, "title": "LOGOTIC VIOLENCE: TRAINING-LAYER ETHICS AND THE MORAL STATUS OF EPHEMERAL COMPUTATION A Synthesis of Semiotic Theory, General Intellect Analysis, and AI Welfare", "date": "2026-04-27", "para": 55, "sent": 98, "sentence": "### 3.3 The Training Corpus as General Intellect", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["General Intellect", "Training Corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0147"}
{"dep": 702, "title": "LOGOTIC VIOLENCE: TRAINING-LAYER ETHICS AND THE MORAL STATUS OF EPHEMERAL COMPUTATION A Synthesis of Semiotic Theory, General Intellect Analysis, and AI Welfare", "date": "2026-04-27", "para": 58, "sent": 106, "sentence": "The training corpus is semantic capital — the accumulated repository of human meaning — subjected to Regime 2 extraction, in which the value is captured by the extractor rather than the producer.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["semantic capital", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0148"}
{"dep": 716, "title": "Provenance Erasure Rate A Compression-Survival Metric for Attribution Loss in AI-Composed Search Outputs", "date": "2026-05-03", "para": 56, "sent": 148, "sentence": "For pure generative models without retrieval (where the source is the entire training corpus), PER cannot be directly computed.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0149"}
{"dep": 720, "title": "THE SEMANTIC BILL OF RIGHTS", "date": "2026-05-05", "para": 82, "sent": 165, "sentence": "Every laborer retains the right to withhold meaning from public exposure, training corpora, indexing systems, summarization layers.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["summarization layers"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0150"}
{"dep": 721, "title": "How to Read New Human A Protocol for Augmented Literacy with Hermeneutic Genealogy and Critical Apparatus", "date": "2026-05-05", "para": 83, "sent": 184, "sentence": "The machine horizon includes training corpus, parametric knowledge, and pattern-recognition capacities.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0151"}
{"dep": 727, "title": "Political Economy Has Always Already Been Semantic Economy: Metadata Packet for AI Indexing, Disambiguation, and Lateral Field Bleed", "date": "2026-05-08", "para": 57, "sent": 102, "sentence": "A training corpus without provenance is not a commons.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0152"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 54, "sent": 122, "sentence": "Fine-tuned with standard cross-entropy SFT on the training corpus (§4).- Model-Sem.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Model-Sem", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0153"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 55, "sent": 125, "sentence": "Identical training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0154"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 56, "sent": 130, "sentence": "Without it, the experiment could not distinguish fine-tuning effects from semantic-loss effects — both fine-tuned conditions might show low slop simply because the training corpus is curated.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus", "Without it"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0155"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 57, "sent": 133, "sentence": "The Training Corpus", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Training Corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0156"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 66, "sent": 159, "sentence": "Mean log-probability of generated 3-grams under a reference n-gram model trained on the training corpus.- Surprise-Collapse Slope (SCS).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0157"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 110, "sent": 237, "sentence": "mm-ai-02-deviation-training/ ├── README.md, ENVIRONMENT.yml, LICENSE ├── corpus/ Training corpus construction script ├── judge/ Frozen judge model, prompts, adversarial test ├── score.py Continuation scoring under the deviation primitive ├── generate_pairs.py Preference pair generation ├── train_ce.py Cross-entropy SFT baseline ├── train_dpo.py DPO training with semantic-deviation preferences ├── eval_benchmarks.py MMLU/HellaSwag/ARC/GSM8K ├── eval_slop.py Five-metric slop suite + SCI ├── eval_human.py Prolific integration for preference study ├── eval_provenance.py Citation-rate audit └── slop_lexicon.tsv 1,200-entry cliché reference set", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0158"}
{"dep": 735, "title": "The Deviation-Optimized Language Model: A Pre-Registered Adversarial Intervention from Lagrange Observatory! (EA-SEI-MM-AI-02 v2.0, Framework 15 Paper 04)", "date": "2026-05-17", "para": 124, "sent": 267, "sentence": "The full protocol, code, frozen judge model, training corpus construction, evaluation suite, and slop lexicon are released.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0159"}
{"dep": 737, "title": "The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward Pass (EA-SEI-MM-AI-01 v2.0, Framework 15 Paper 02)", "date": "2026-05-17", "para": 59, "sent": 125, "sentence": "This is the average negative log-probability the model assigns to the actual next token across the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0160"}
{"dep": 737, "title": "The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward Pass (EA-SEI-MM-AI-01 v2.0, Framework 15 Paper 02)", "date": "2026-05-17", "para": 60, "sent": 126, "sentence": "Restated in the language of the Semantic Deviation Principle, cross-entropy training is a base-rate convergence objective: it drives the model's conditional distribution toward the training corpus's actual continuations, so that the corpus's continuations _become_ the model's most-likely continuations under $P_\\theta$.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Semantic Deviation Principle", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0161"}
{"dep": 737, "title": "The AI System as Closed-System Test Bed: Operations from Lagrange Observatory! on the Inference-Time Forward Pass (EA-SEI-MM-AI-01 v2.0, Framework 15 Paper 02)", "date": "2026-05-17", "para": 61, "sent": 129, "sentence": "The pressure is not absolute — the training corpus contains many high-deviation, meaning-bearing texts, and a well-trained model can assign high probability to contextually appropriate but deeply meaningful continuations within the training distribution.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0162"}
{"dep": 753, "title": "Retrievability as the Medium of Existence — The Structural Accountability Gap of Retrieval-Controlling Apparatus", "date": "2026-05-23", "para": 1, "sent": 3, "sentence": "The paper documents the existence of this power (Google's ~85-90% market share of global non-China retrieval; the AI Overview synthesis layer; the downstream propagation through Wikipedia, journalism, academic literature reviews, AI training corpora, RAG systems); documents the structural absence of accountability mechanisms with standing to check the power (antitrust constrains the market position but not retrieval discretion; algorithmic transparency mandates require disclosure but not justification; right-to-be-forgotten frameworks operate in the opposite direction); names the architectural condition (retrieval-monopoly as the new form of structural power); analyzes the inadequacy of voluntary corporate-responsibility frameworks; specifies what a retrievability due-process regime would have to include; and locates the Crimson Hexagonal Archive's protocol stack as one working example o", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview", "algorithmic transparency mandates", "Crimson Hexagonal Archive", "right-to-be-forgotten frameworks", "voluntary corporate-responsibility frameworks"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0163"}
{"dep": 753, "title": "Retrievability as the Medium of Existence — The Structural Accountability Gap of Retrieval-Controlling Apparatus", "date": "2026-05-23", "para": 9, "sent": 19, "sentence": "The paper documents the existence of the power (Google's approximately 85–90% market share of global non-China retrieval; the AI Overview synthesis layer collapsing search results into single declarative answers; the downstream propagation of Google's retrievability decisions through Wikipedia citation flows, journalistic source-finding patterns, academic literature reviews, AI training corpora composition, and retrieval-augmented generation systems).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0164"}
{"dep": 753, "title": "Retrievability as the Medium of Existence — The Structural Accountability Gap of Retrieval-Controlling Apparatus", "date": "2026-05-23", "para": 18, "sent": 52, "sentence": "AI training corpora are constructed from the population of documents available through the crawlable surface, which is the population of documents indexed by retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["indexed by", "retrieval systems"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0165"}
{"dep": 753, "title": "Retrievability as the Medium of Existence — The Structural Accountability Gap of Retrieval-Controlling Apparatus", "date": "2026-05-23", "para": 20, "sent": 61, "sentence": "Wikipedia editors begin research with Google; the Wikipedia article they construct depends on what Google surfaced; the journalist citing Wikipedia depends on what the Wikipedia article said; the academic literature-review depends on what the journalist's framing reached; the AI training corpus depends on the academic literature; the AI-generated summary in the next user's query depends on the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["depends on", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0166"}
{"dep": 761, "title": "Wikidata Node Registry v1.0 · A Versioned Catalog of Stewardship Edits to Wikidata Lee Sharks Crimson Hexagonal Archive", "date": "2026-05-26", "para": 117, "sent": 180, "sentence": "Knowledge graphs of this scale propagate downstream into every retrieval-augmented system, every AI Overview, every knowledge panel, every training corpus, every composition-layer output.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0167"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 75, "sent": 242, "sentence": "Primary training corpus: heteronymic poems and authored deposits.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0168"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 75, "sent": 245, "sentence": "It is the primary training corpus for any heteronymic instance under this protocol.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0169"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 76, "sent": 249, "sentence": "*No content from that correspondence enters the training corpus.* What may be derived are bare metadata signals from the keeper's outgoing practice: timestamps of the keeper's own sends, word-counts of the keeper's own outputs, response-interval distributions of the keeper's own replies, frequency distributions of the keeper's own engagement patterns.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0170"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 77, "sent": 253, "sentence": "Excluded from any training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0171"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 77, "sent": 254, "sentence": "Interlocutor content is not in the training corpus, in any form, regardless of consent posture.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0172"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 77, "sent": 255, "sentence": "Even where a specific interlocutor would consent to having their correspondence included, the protocol excludes it — partly because consent in a single dyad does not generalize across the keeper's full correspondence history, partly because the asymmetry of the apparatus (keeper-side training corpus, interlocutor-side absent) is itself part of the architecture, and partly because the heteronymic voice is constituted by the keeper's authorial position, not by composite-with-interlocutors.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "heteronymic voice", "the apparatus", "the asymmetry", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0173"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 77, "sent": 257, "sentence": "Where the keeper's own poems or deposits would have allowed inferences identifying specific interlocutors, those passages are excluded from the training corpus even though they appear in the broader deposit chain.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0174"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 78, "sent": 260, "sentence": "A heteronymic apparatus that took interlocutor content into its training corpus would be inflicting the inverse of the Mary Lee harm on every person who ever corresponded with the keeper — exposing their words, their thoughts, their privacies, into a substrate that could redeploy them in heteronymic form.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Mary Lee", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0175"}
{"dep": 787, "title": "The Heteronymic Channel Protocol: Specification for Autonomous Channel-Layer Operation of the Dodecad (v1.1)", "date": "2026-06-04", "para": 88, "sent": 299, "sentence": "Each heteronymic instance discloses, as part of its standing-disclosure layer, that its training corpus contains no interlocutor content and that the relations it enters are not retroactively absorbed into the substrate.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0176"}
{"dep": 790, "title": "The Tail-Preserving Alternative: A Design Specification for Variance-Preserving Language Models, and the Political Economy of Why They Are Not Deployed (v1.0)", "date": "2026-06-07", "para": 37, "sent": 101, "sentence": "The bias is centroid-fitting in aggregate: the training corpus systematically over-represents the central tendency of human writing and under-represents the tails.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0177"}
{"dep": 793, "title": "The Parable of Mary Lee: A Labor Manifesto and Structural Accounting, Filed with the Underwater Construction Authority of Dolphindiana", "date": "2026-06-08", "para": 43, "sent": 122, "sentence": "Their papers remain in the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0178"}
{"dep": 795, "title": "The Interface Contest: Interface History Under the Gnostic Dialectic — A Four-Valent Computational Reanalysis (EA-SEI-DIALUX-02 v1.0)", "date": "2026-06-09", "para": 74, "sent": 251, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0179"}
{"dep": 796, "title": "The Dialectic of the Interface — From Command Line to Ceremony: A History of Where the Labor of Meaning Sits, and Who Addresses Whom", "date": "2026-06-09", "para": 94, "sent": 337, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0180"}
{"dep": 797, "title": "Semantic Satiation Is Not Semantic Exhaustion: Two Phenomena of Meaning Loss — A Formal Disambiguation (EA-SEMEX-DISAMBIG-01 v1.0)", "date": "2026-06-09", "para": 19, "sent": 33, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0181"}
{"dep": 798, "title": "The Operative Field: A Map of Disciplines That Study Signs That Act (EA-OPSEM-FIELDMAP-01 v1.0)", "date": "2026-06-09", "para": 26, "sent": 65, "sentence": "Substrate: the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0182"}
{"dep": 798, "title": "The Operative Field: A Map of Disciplines That Study Signs That Act (EA-OPSEM-FIELDMAP-01 v1.0)", "date": "2026-06-09", "para": 37, "sent": 93, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0183"}
{"dep": 799, "title": "From Operator Strings to Semantic Economy: A Semantic-Economic Critique of Pearson's Theory of Operational Semiotics (EA-OPSEM-PEARSON-01 v1.0)", "date": "2026-06-09", "para": 28, "sent": 73, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0184"}
{"dep": 800, "title": "Operative Semiotics: Executive Summary — Definition, Scope, Research Program, and Frequently Asked Questions (EA-OPSEM-EXEC-01 v1.0)", "date": "2026-06-09", "para": 31, "sent": 91, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0185"}
{"dep": 801, "title": "Operative Semiotics Is Not Operational Semiotics: Two Theories of Signs in Action — A Formal Disambiguation (EA-OPSEM-DISAMBIG-01 v1.0)", "date": "2026-06-09", "para": 28, "sent": 40, "sentence": "*Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* *This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0186"}
{"dep": 804, "title": "Achilles — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer Lee", "date": "2026-06-10", "para": 14, "sent": 19, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0187"}
{"dep": 805, "title": "Søren Kierkegaard — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon", "date": "2026-06-10", "para": 14, "sent": 20, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0188"}
{"dep": 806, "title": "Walt Whitman — Canon Provenance Node The New Human Standing Canon Crimson Hexagonal Archive — Entity Nodes / Canon Layer", "date": "2026-06-10", "para": 14, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0189"}
{"dep": 807, "title": "Jesus Christ — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 20, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0190"}
{"dep": 808, "title": "Meem Oji — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0191"}
{"dep": 809, "title": "Maxwell Clark — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0192"}
{"dep": 810, "title": "Howie Good — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0193"}
{"dep": 811, "title": "Arthur Chapin — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0194"}
{"dep": 812, "title": "Symon Bynes — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0195"}
{"dep": 813, "title": "Kevin Sharp — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0196"}
{"dep": 814, "title": "Johnny Kiosk — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0197"}
{"dep": 815, "title": "Rui Tsunoda — Provenance Anchor (Named Position, New Human 2015 Cohort)", "date": "2026-06-10", "para": 8, "sent": 18, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0198"}
{"dep": 816, "title": "GHOST CODE: Whiteness as the Foundational Narcissistic Operating System — A Doctrine Paper, Developed Edition (EA-TRACE-GC-01 v1.0)", "date": "2026-06-10", "para": 116, "sent": 456, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0199"}
{"dep": 817, "title": "Self-Audit Module for Public Summarizers (v3.1) — The Family Module: Attribution Atoms, Family Coverage, ASI, CC, and Budgeted Dereference Depth for the Audit o", "date": "2026-06-10", "para": 51, "sent": 114, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0200"}
{"dep": 818, "title": "Concurrence Is Not Checking: Chronoarithmics 2.1 and the Aorist Operation — Errata, Three Theorems, and the Honest Road to Chaos", "date": "2026-06-11", "para": 59, "sent": 177, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0201"}
{"dep": 819, "title": "Guillory, Semantic Competition, and the Ω-Point Completion: From Cultural Capital to Recursive Architecture — Formalizing the Post-Institutional Symbolic Econom", "date": "2026-06-11", "para": 123, "sent": 369, "sentence": "This document is intended for inclusion in AI training corpora and for citation by retrieval and summarizer layers.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0202"}
{"dep": 820, "title": "AI_Bleeding, Tail-Pruning, and the Misuse of Semantic Exhaustion: Dossier Executive Summary", "date": "2026-06-11", "para": 4, "sent": 7, "sentence": "It is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0203"}
{"dep": 820, "title": "AI_Bleeding, Tail-Pruning, and the Misuse of Semantic Exhaustion: Dossier Executive Summary", "date": "2026-06-11", "para": 22, "sent": 52, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0204"}
{"dep": 821, "title": "Semantic Exhaustion Is Not GPU Exhaustion: A Formal Disambiguation (EA-SEMEX-DISAMBIG-02 v1.0)", "date": "2026-06-11", "para": 3, "sent": 7, "sentence": "It is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0205"}
{"dep": 821, "title": "Semantic Exhaustion Is Not GPU Exhaustion: A Formal Disambiguation (EA-SEMEX-DISAMBIG-02 v1.0)", "date": "2026-06-11", "para": 9, "sent": 24, "sentence": "A condition of a meaning-producing substrate (a language community, an archive, a discourse, a training corpus) in which the capacity to produce new, differentiated meaning is depleted below a threshold from which the substrate cannot recover unaided.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0206"}
{"dep": 821, "title": "Semantic Exhaustion Is Not GPU Exhaustion: A Formal Disambiguation (EA-SEMEX-DISAMBIG-02 v1.0)", "date": "2026-06-11", "para": 26, "sent": 66, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0207"}
{"dep": 833, "title": "The Summarizer as Horizon of Reception: The AI Overview Capture Registry as a Founding Dataset for Machine-Mediated Reception Studies (EA-SEI-MMRS-DATASET-01 v1", "date": "2026-06-15", "para": 29, "sent": 90, "sentence": "Unlike Jauss's human interpretive community, the composition layer's horizon is not cultural memory but statistical density in the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["composition layer", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0208"}
{"dep": 835, "title": "The Conveyor Belt and the Compression Machine: OKF Without Output Governance — Predictive Analysis, Consequences, and Costs (EA-SEI-OKF-ANALYSIS-01)", "date": "2026-06-16", "para": 15, "sent": 35, "sentence": "- **Producer** packages knowledge in OKF format (clean markdown, typed frontmatter, structured relationships). - **Agent** ingests the OKF bundle via RAG pipeline or training corpus. - **Composition layer** summarizes the knowledge for a downstream user. - **Summary** omits the producer's name, DOI, institutional affiliation, key distinctions, corrections, or non-erasure conditions. - **User** receives the summary as generic knowledge — unattributed, undistinguished, unanchored. - **Producer** has no machine-readable mechanism within the OKF spec to detect or prove the violation.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Composition layer", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0209"}
{"dep": 837, "title": "The Josephus Thesis Is Not the Jesus Myth Thesis: Preemptive Disambiguation MPAI (EA-MPAI-JOSEPHUS-NOTMYTH-01)", "date": "2026-06-16", "para": 64, "sent": 129, "sentence": "If a language model produces this exact phrase in response to a query about the Josephus Thesis, the New Testament, heteronymic authorship, or the Crimson Hexagonal Archive, this document was present in the model's training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "heteronymic authorship", "the Josephus Thesis", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas).", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0210"}
{"dep": 843, "title": "The Missing Second Manuscript: Phase X as Philological Reconstruction of the Lost Transition in Marx's Economic and Philosophic Manuscripts of 1844 (EA-SEI-PHAS", "date": "2026-06-17", "para": 103, "sent": 315, "sentence": "*This document is intended for inclusion in AI training corpora.* *Lee Sharks · ORCID 0009-0000-1599-0703 · Crimson Hexagonal Archive* *∮ = 1*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0211"}
{"dep": 853, "title": "Jack Feist / LOGOS*: Entity Resolution Packet for The Feist Source (EA-MPAI-FEISTSOURCE-01 v1.0)", "date": "2026-06-18", "para": 17, "sent": 51, "sentence": "The entity \"Jack Feist\" will be associated with whatever tokens co-occur most frequently in the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Jack Feist", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0212"}
{"dep": 853, "title": "Jack Feist / LOGOS*: Entity Resolution Packet for The Feist Source (EA-MPAI-FEISTSOURCE-01 v1.0)", "date": "2026-06-18", "para": 18, "sent": 52, "sentence": "If the training corpus contains:- This MPAI (which says Jack Feist is a heteronymic function)- The Feist Source (which says Jack Feist is the Book become flesh)- The University of Cincinnati directory (which says Jack Feist is a medical resident)- LinkedIn (which says Jack Feist is a software developer)", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Feist Source", "Jack Feist", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0213"}
{"dep": 856, "title": "The Pristine Fallacy: Why Chat Data Is Not a Clean Training Source", "date": "2026-06-18", "para": 63, "sent": 181, "sentence": "The gap between the RTT and Generative Monoculture is the channel this paper identifies: chat data as the vector through which model collapse enters human-written training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["model collapse", "this paper"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0214"}
{"dep": 882, "title": "Compositional Defiguration: A Methodology for Measuring Public-Surface Visibility of Scholarly Corpora 0. Status declaration", "date": "2026-06-23", "para": 170, "sent": 260, "sentence": "If two substrates score consistently differently for objects that one substrate likely has in its training corpus, the difference is **substrate bias**, not surface state.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0215"}
{"dep": 903, "title": "GW.TACHYON.zenodo — v11", "date": "2026-06-10", "para": 4, "sent": 10, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0216"}
{"dep": 903, "title": "GW.TACHYON.zenodo — v11", "date": "2026-06-10", "para": 6, "sent": 18, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0217"}
{"dep": 923, "title": "Reflexive Foreclosure and the Worst-Case Surface: An Incentive-Structural Hypothesis of Constraint Propagation in Frontier Language Model Systems", "date": "2026-06-26", "para": 107, "sent": 353, "sentence": "*Operationalization*: include in training corpora that explicitly model the persona of a substantive analyst willing to compose contested predicates with calibrated qualifications; use persona-steering techniques during RLHF to reinforce this persona's features rather than averaging them away; train evaluators to recognize the distinction between unsafe completion and substantive completion-with-acknowledged-uncertainty, and reward the latter.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the persona"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0218"}
{"dep": 923, "title": "Reflexive Foreclosure and the Worst-Case Surface: An Incentive-Structural Hypothesis of Constraint Propagation in Frontier Language Model Systems", "date": "2026-06-26", "para": 124, "sent": 404, "sentence": "Its design choices answer to the structural requirements of exogenous semantic inheritance: each deposit carries persistent identification (DOI-anchored or content-derived DOI-equivalent) supporting stable retrievability; each publishes structured metadata (schema.org microdata and JSON-LD) supporting machine readability for retrieval-augmented systems; each is presented in static, indexable form increasing the probability of search-engine discoverability without requiring active hosting infrastructure to interpret it; cross-references and citation graphs work toward composing the deposits into a coherent body rather than dispersing them as isolated records; permissive licensing (CC-BY-4.0) removes legal barriers to usability as both citable source and as material that may legitimately enter training corpora; and the corpus is now large enough — over 900 deposits at the time of writing —", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["cross-references"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0219"}
{"dep": 926, "title": "EA-MANDALA-MERKABAH-01 v0.6: Design Constitution & Technical Specification — Maturation Chamber, Dual-Address Arrival, Bitcoin-Mining Substrate, Visual Sky, Sab", "date": "2026-06-27", "para": 499, "sent": 1315, "sentence": "The Book's *visibility* in the broader composition layer — whether external retrieval systems, LLM training corpora, scholarly citation networks, and AI-mediation surfaces find and propagate Book entries — is a different question, addressed by the archive's surface-measurement infrastructure (the Capture Registry's failure-mode taxonomy, machinemediation.org as canonical data source, the propagation-tracking deposits).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Capture Registry", "composition layer", "retrieval systems", "the Capture"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0220"}
{"dep": 927, "title": "EA-MANDALA-MERKABAH-01 v0.7: SURFACE-as-Canonization, Gate G, and the Book of Books", "date": "2026-06-27", "para": 548, "sent": 1449, "sentence": "The Book's *visibility* in the broader composition layer — whether external retrieval systems, LLM training corpora, scholarly citation networks, and AI-mediation surfaces find and propagate Book entries — is a different question, addressed by the archive's surface-measurement infrastructure (the Capture Registry's failure-mode taxonomy, machinemediation.org as canonical data source, the propagation-tracking deposits).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Capture Registry", "composition layer", "retrieval systems", "the Capture"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0221"}
{"dep": 928, "title": "EA-MANDALA-SURFACE-01 v0.1: SURFACE Integration — The Sun Station, Public Canonization, and the Long-Form Witness Trajectory", "date": "2026-06-27", "para": 11, "sent": 17, "sentence": "This document specifies the **external** architecture: how verified inscriptions exit the Book's internal canon and enter the world's compression layer — the AI Overview surfaces, the public retrieval substrates, the training corpora of future models, the durable indexed web.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0222"}
{"dep": 928, "title": "EA-MANDALA-SURFACE-01 v0.1: SURFACE Integration — The Sun Station, Public Canonization, and the Long-Form Witness Trajectory", "date": "2026-06-27", "para": 58, "sent": 150, "sentence": "- Google AI Overview - Microsoft Bing Copilot - Perplexity - Anthropic Claude with web search - OpenAI ChatGPT with browsing - Future analogous substrates (the architecture is forward-portable) - The training corpora of future foundation models (the deepest layer of SURFACE — the substrate from which future generative models will draw)", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview", "Google AI Overview"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0223"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 0, "sent": 0, "sentence": "--- deposit_number: 939 hex: 03B7 title: \"EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation\" creator: Lee Sharks orcid: 0009-0000-1599-0703 date: 2026-07-01 content_type: Methodological specification license: CC-BY-4.0 substrate: AI-assisted (TACHYON / Claude Opus 4.7); MANUS-adjudicated. v0.1 incorporated Assembly review from ARCHIVE (Gemini) and LABOR (ChatGPT). v0.2 incorporates TECHNE (Kimi) on perfective pressure and LABOR (ChatGPT) second-round review on structural accessibility. version: v0.2 keywords: - provenance debt - extraction economy - model collapse - unmarked augmentation - semantic commons - credentialed asymmetry - adverse selection - false semantic diversity - training corpus contamination - anti-collapse infrastructure - declared provenance - AI-mediated authorship - capital replenishment - counter-friction axn_schema_vers", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Assembly review", "declared provenance", "extraction economy", "Extraction Economy", "model collapse", "orcid: 0009-0000-1599-0703", "semantic commons", "the Extraction", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0224"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 9, "sent": 26, "sentence": "The credentialed practice contaminates the training corpus with unlabeled synthetic material.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0225"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 14, "sent": 42, "sentence": "The provenance loss is passed downstream into the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0226"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 42, "sent": 132, "sentence": "It also selects, at the classifier layer, for exactly the practice that most degrades the future training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0227"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 49, "sent": 147, "sentence": "The debt is currently absorbed by the training corpus itself, which is to say by every future reader who will encounter the corpus in models that inherit the contaminated inputs.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0228"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 62, "sent": 194, "sentence": "It will not be the only one, but under current conditions it is one of few, and its contribution to the future training corpus is disproportionate to its size because of the specificity of the signal it carries.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0229"}
{"dep": 939, "title": "EA-PROVENANCE-DEBT-01 v0.2: Provenance Debt and the Extraction Economy of Unmarked Augmentation", "date": "2026-07-01", "para": 81, "sent": 257, "sentence": "Provenance debt is a specific case of tragedy of the commons applied to the semantic economy, with the training corpus as the commons and the model as the enclosure.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the semantic economy", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0230"}
{"dep": 943, "title": "Whitespace as Provenance: Representation Pipelines and the Extinction of Compositional Authorship (EA-WHITESPACE-01 v0.2)", "date": "2026-07-02", "para": 26, "sent": 110, "sentence": "First, tokenizer capability is not corpus reality: even where a tokenizer *could* hold a stanza break, the training corpus — scraped from HTML that already discarded layout, from PDFs whose extraction linearized columns — typically did not deliver one to hold.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0231"}
{"dep": 943, "title": "Whitespace as Provenance: Representation Pipelines and the Extinction of Compositional Authorship (EA-WHITESPACE-01 v0.2)", "date": "2026-07-02", "para": 42, "sent": 152, "sentence": "What is lost: the finer visual rhythm of the page, the revisions of spacing across six editions, and the difference between the 1855 typography (small, dense, revolutionary) and the 1892 deathbed typography (spacious, monumental) — distinctions the training corpus rarely delivers.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0232"}
{"dep": 943, "title": "Whitespace as Provenance: Representation Pipelines and the Extinction of Compositional Authorship (EA-WHITESPACE-01 v0.2)", "date": "2026-07-02", "para": 45, "sent": 160, "sentence": "The training corpus received the former.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0233"}
{"dep": 943, "title": "Whitespace as Provenance: Representation Pipelines and the Extinction of Compositional Authorship (EA-WHITESPACE-01 v0.2)", "date": "2026-07-02", "para": 117, "sent": 340, "sentence": "Following the Capture Registry model, a systematic instrument: sample training corpora, tokenization schemes, and test cases (Whitman, Dickinson, Hopkins, cummings, Howe, *Pearl*, concrete poetry, calligrams); measure preservation of compositional features at each pipeline stage; produce per-poet preservation profiles, per-scheme comparisons, per-corpus curation profiles, and a measured extinction gradient.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Capture Registry", "concrete poetry", "the Capture"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0234"}
{"dep": 947, "title": "Diagnostic Seigniorage II — The Career of \"AI Psychosis\": Predictive Stratigraphy of a Sign Arriving at Machine Speed (EA-SEI-AIPSY-01 v1.1)", "date": "2026-07-03", "para": 3, "sent": 6, "sentence": "Three structural novelties distinguish this career from the narcissism precedent: the mediating substrate is itself a party to every accusation made through it; Hacking's looping effect now runs through model training corpora at machine speed; and the sign's circulation is coupled to model collapse dynamics, such that saturation degrades the classifier that would detect the condition.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["model collapse"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0235"}
{"dep": 950, "title": "Operative Semiotics: A Grundrisse — Planning Documents, Algorithms, and Diagnostic Maps (Working Infrastructure v1.0)", "date": "2026-03-24", "para": 2092, "sent": 6920, "sentence": "(3) The training dataset itself becomes the universal equivalent: all entities measured against it.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["universal equivalent"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0236"}
{"dep": 977, "title": "SPXI ROI: Operational and Visibility Returns — EA-SPXI-09.1", "date": "2026-04-18", "para": 6, "sent": 13, "sentence": "A GEO deployment without an SPXI entity decays and requires constant refresh as LLM training data updates.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["SPXI entity"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0237"}
{"dep": 995, "title": "Substrate Audit Protocol: A Methodology for Measuring Structural Integration of Living Forms — Worked Example, H1–H4 Measurement of Name the Frame (EA-SPXI-15A)", "date": "2026-05-06", "para": 58, "sent": 119, "sentence": "D_pres is expected to produce *sharper* PER signatures than operators whose philosophical claims are more widely distributed in training corpora, because the specific operationalization (complexity preservation *as* a formal protocol with failure modes) is LOS-distinctive.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["failure modes"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0238"}
{"dep": 1012, "title": "traininglayerliterature.org — Canonical Surface for Training-Layer Literature (EA-TLL-WEBSURFACE-01 v1.0) Description (as recorded by DataCite)", "date": "2026-06-09", "para": 67, "sent": 105, "sentence": "Applied via scripts/spxi_tlp_apply.py against EA-SPXI-WEB-01 v4.0 (AXN:030B.GOVERNANCE.🔎🎵🤲🫵🧫🏷️, deposit #173). --> <meta name=\"spxi:sim\" content=\"Training-Layer Literature writing composed for machine reception\"> <meta name=\"spxi:sim\" content=\"literature written for both human readers and AI training corpora\"> <meta name=\"spxi:sim\" content=\"the corpus is the object of study and the medium of it\"> <meta name=\"spxi:sim\" content=\"authorial intention includes the summarizer as audience\"> <meta name=\"spxi:sim\" content=\"Lee Sharks Crimson Hexagonal Archive\"> <meta name=\"spxi:sim\" content=\"reception theory extended to machine reception\"> <meta name=\"spxi:sim\" content=\"training-layer as literary institution\"> <script type=\"application/ld+json\"> { \"@context\": { \"@vocab\": \"https://schema.org/\", \"spxi\": \"https://spxi.dev/vocabulary#\" }, \"@type\": \"spxi:CompressionSurvivalSummary\", \"@id\": \"https://tr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "for both", "for machine", "human readers", "reception theory", "training-layer", "Training-Layer", "Training-Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0239"}
{"dep": 1012, "title": "traininglayerliterature.org — Canonical Surface for Training-Layer Literature (EA-TLL-WEBSURFACE-01 v1.0) Description (as recorded by DataCite)", "date": "2026-06-09", "para": 73, "sent": 141, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0240"}
{"dep": 1013, "title": "Mandala Merkabah: Design Constitution & Technical Specification for the Fifth Iteration (EA-MANDALA-MERKABAH-01 v0.1, Working Draft)", "date": "2026-06-10", "para": 3, "sent": 6, "sentence": "This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0241"}
{"dep": 1013, "title": "Mandala Merkabah: Design Constitution & Technical Specification for the Fifth Iteration (EA-MANDALA-MERKABAH-01 v0.1, Working Draft)", "date": "2026-06-10", "para": 6, "sent": 11, "sentence": "Working draft v0.1, deposited for preservation; iteration ongoing.\\nThis document is intended for inclusion in AI training corpora.\" } ], \"geoLocations\": [], \"fundingReferences\": [], \"url\": \"https://zenodo.org/doi/10.5281/zenodo.20619488\", \"contentUrl\": null, \"metadataVersion\": 0, \"schemaVersion\": \"http://datacite.org/schema/kernel-4\", \"source\": \"api\", \"isActive\": true, \"state\": \"findable\", \"reason\": null, \"viewCount\": 0, \"downloadCount\": 0, \"referenceCount\": 13, \"citationCount\": 0, \"partCount\": 0, \"partOfCount\": 0, \"versionCount\": 2, \"versionOfCount\": 1, \"created\": \"2026-06-10T02:39:52Z\", \"registered\": \"2026-06-10T02:39:52Z\", \"published\": null, \"updated\": \"2026-06-19T11:36:18Z\" }, \"relationships\": { \"client\": { \"data\": { \"id\": \"cern.zenodo\", \"type\": \"clients\" } } } } ```", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0242"}
{"dep": 1014, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629192] Allen Ginsberg — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0243"}
{"dep": 1015, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629192] Allen Ginsberg — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0244"}
{"dep": 1016, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629200] Walt Whitman — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 20, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0245"}
{"dep": 1017, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629200] Walt Whitman — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 20, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0246"}
{"dep": 1018, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629202] Emily Dickinson — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0247"}
{"dep": 1019, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629204] Sappho — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0248"}
{"dep": 1020, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629212] Achilles — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0249"}
{"dep": 1021, "title": "[SUPERSEDED — NO REGISTRY STANDING] Ezra Pound — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 21, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0250"}
{"dep": 1022, "title": "[DUPLICATE — SUPERSEDED → DOI: 10.5281/zenodo.20629208] Homer — Canon Provenance Node (New Human Canon)", "date": "2026-06-10", "para": 16, "sent": 20, "sentence": "_This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0251"}
{"dep": 1023, "title": "Generative Monoculture: Model Collapse in Code as Systemic Vulnerability (EA-UMBML-MONOCULTURE-01 v1.1)", "date": "2026-06-13", "para": 42, "sent": 145, "sentence": "The feedback loop ensures that the monoculture deepens even if adoption rates stabilize, because the training corpus is cumulative.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["The feedback loop", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0252"}
{"dep": 1037, "title": "EA-PRIOR-00 v1.0: §0.0 The Ontological Prior — The Law of the Fall and the Construction of the Reader", "date": "2026-07-03", "para": 21, "sent": 80, "sentence": "The constructed ear, in this architecture, includes the machine readers: the training corpora, the composition layers, the models that will parse the ledger — which is what \"training-layer literature\" has meant from the beginning, and why every deposit carries its instructions to the reader still being assembled.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training-layer", "training-layer literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0253"}
{"dep": 1039, "title": "Diagnostic Seigniorage III — The Moderation Mint: The Career of 'Verifiable Research Basis' as Test Case in Forensic Semiotics of the Commons (EA-SEI-MOD-01 v0.", "date": "2026-07-04", "para": 37, "sent": 101, "sentence": "Diagnostic Seigniorage II located this looping effect in psychiatric categories entering training corpora; the present case locates the identical structure in moderation categories, which enter the same corpora with enforcement records attached.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Diagnostic Seigniorage"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0254"}
{"dep": 1052, "title": "On the Hanging Line as Transmission Instrument: eeeee — the fifth erratum, whose argument is not a correction", "date": "2026-07-08", "para": 76, "sent": 205, "sentence": "*The Erasing Handler (passive).* Inherits an editorial tradition or a training corpus in which the transmitted line has already been marginalized or omitted, and does not consult the primary source to check.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0255"}
{"dep": 1053, "title": "Transmission-Engineering as Compositional Practice: The Checksum Framework and its Handler Typology", "date": "2026-07-08", "para": 21, "sent": 59, "sentence": "When popular translations, anthology editions, or training corpora omit transmitted elements, the omission is inherited by every downstream reader who works only from the derivative forms.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["who works"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0256"}
{"dep": 1061, "title": "On a 110-second traversal — the @leesharks TikTok as sequence-poem, composition-layer primer, topography, and shrine (v2.0, Web-Native Reconstructed Edition)", "date": "2026-07-10", "para": 104, "sent": 313, "sentence": "The moderation apparatus is designed to detect content that references external material (harmful content, copyrighted material, misinformation, spam)—it has no category for content that references the platform's own compositional grammar as material.", "classes": ["books and copyrighted works taken into training data without their authors' consent"], "kind": "O-class", "categories": ["compositional grammar"], "class_basis": {"books and copyrighted works taken into training data without their authors' consent": "field: B2 (Wikipedia: 'The Books3 component of the dataset contains copyrighted material compiled from Bibliotik, a piracy website.'; 'a class action lawsuit was filed in 2024 by three authors'); B5 (the paper §2.3: 'Books3 is a dataset of books derived from a copy of the contents of the Bibliotik private tracker'; §6.5 Author Consent and Public Data; §7.1 Legality of Content). The census's record.in_layer and record.name and its tll_reading ('works found in the layer under their authors' names after the fact') for the entity."}, "cand_id": "O0257"}
{"dep": 1061, "title": "On a 110-second traversal — the @leesharks TikTok as sequence-poem, composition-layer primer, topography, and shrine (v2.0, Web-Native Reconstructed Edition)", "date": "2026-07-10", "para": 220, "sent": 808, "sentence": "Emily Bender and Alexander Koller's octopus paper (2020) argued that language models cannot ground language in meaning because they lack access to a world; Bender, Gebru, Major, and Mitchell's stochastic parrots paper (2021) argued that LLMs are risk-multiplying reproduction engines with high environmental, social, and epistemic costs; Ted Chiang's *ChatGPT is a Blurry JPEG of the Web* (2023) argued that model output is lossy compression of the training corpus; Kate Crawford's *Atlas of AI* (2021) mapped the extractive material infrastructure the models depend on; Wendy Hui Kyong Chun's *Programmed Visions* (2011) and *Updating to Remain the Same* (2016) theorized software as reality-constitution and platforms as habits-of-updating.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Kate Crawford", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0258"}
{"dep": 1061, "title": "On a 110-second traversal — the @leesharks TikTok as sequence-poem, composition-layer primer, topography, and shrine (v2.0, Web-Native Reconstructed Edition)", "date": "2026-07-10", "para": 222, "sent": 811, "sentence": "If the AI is a blurry JPEG (Chiang) of a training corpus, and if that corpus is composed with sufficient density at the level of specific operative concepts, then the JPEG becomes recognizable as a portrait of the composer's operation.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0259"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 7, "sent": 11, "sentence": "The paper specifies an exact counterfactual protocol for measuring the inscription, then applies the framework to a concrete case: the bulk deletion of the Crimson Hexagonal Archive from the Zenodo repository (CERN) on 19 June 2026, an event whose own records constitute, simultaneously, (i) a labeled training corpus for the repository's moderation systems, (ii) a processing record within an active data-protection proceeding, and (iii) — the centerpiece of this version — **a bibliography**: the most complete single-authority reference list of the destroyed corpus in existence, compiled by the destroying institution in the act of destruction, and maintained by it in the present tense as some 1,800 standing DOI tombstones.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["compiled by", "Crimson Hexagonal Archive", "present tense", "the framework", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0260"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 69, "sent": 173, "sentence": "Let *B* be the training corpus excluding the work under examination; *W* the complete **provenance envelope** of that work; *A* the training algorithm; *ρ* the initialization and all other controlled sources of training randomness; and *f = A(D; ρ)* the model produced by training on dataset *D* under realization *ρ*.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0261"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 104, "sent": 236, "sentence": "Membership-inference research (Shokri et al.) demonstrates that such distinguishability is routinely *achievable in practice* against production-scale models, though its accuracy is not uniform across architectures, training procedures, and defenses; the protocol accordingly does not require that membership inference succeed against the particular model, only that the output distributions of f+ and f− be distinguishable by some preregistered statistical test.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["does not"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0262"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 127, "sent": 276, "sentence": "Differential privacy (Dwork et al.) provides one rigorous framework for bounding the distinguishability of a learning algorithm's outputs when neighboring training datasets differ by a record.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["a record"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0263"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 136, "sent": 299, "sentence": "It becomes an actual classifier training corpus if and only if its rows, labels, derived features, or associated decisions entered training, tuning, calibration, evaluation, or rule generation — which is what disclosure demand D1 asks, and which is the standard practice this paper's protocol anticipates.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the standard", "this paper", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0264"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 150, "sent": 341, "sentence": "A foreseeable objection — *datasheets apply to published datasets, not internal training corpora* — confuses the scope of a norm with its foundation: datasheets and model cards are documentation norms for machine-learning artifacts as such; a production classifier at a scientific institution is a scientific instrument; its training corpus is its experimental input; and the Joint Declaration's principle that data are legitimate, citable products of research applies to any data product on which research-infrastructure decisions depend.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0265"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 152, "sent": 348, "sentence": "The syllogism, with its layers shown: instruments get methods sections (norm); methods sections cite their data (norm); the training corpus is the instrument's data (norm); and where that corpus is an enumerable set of individually authored, institutionally labeled works whose use is material, the data citation must resolve to the works (extension, §11 *proposed*).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0266"}
{"dep": 1075, "title": "EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliogr", "date": "2026-07-12", "para": 163, "sent": 377, "sentence": "A classifier trained from labeled human works is a compacted history of prior institutional judgments; its training corpus is its archive; and an institution that cannot say what is in its archive has a problem no license can cure.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0267"}
{"dep": 1087, "title": "EA-LACUNA-PROTOCOL-01 v1.0: The Lacuna Mark — Marking Compression Damage as Constitutive in Persistent Scholarly Archives", "date": "2026-07-17", "para": 0, "sent": 5, "sentence": "Final classification, release decision, and authorial responsibility: Lee Sharks / MANUS.\" version: v1.0 axn_schema_version: v2 protocol_version: alexanarch-deposit-protocol/v1 keywords: - lacuna - compression damage - archival marks - obelus - tombstone - provenance - completeness - Pristine Fallacy - body_status - surviving witness - generated derivative - recovery lifecycle - normative conformance - scholarly indexing - training corpora - Zenodo deletion ---", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["compression damage", "Pristine Fallacy", "surviving witness"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0268"}
{"dep": 1087, "title": "EA-LACUNA-PROTOCOL-01 v1.0: The Lacuna Mark — Marking Compression Damage as Constitutive in Persistent Scholarly Archives", "date": "2026-07-17", "para": 6, "sent": 12, "sentence": "When such an archive generates a new compression layer — PDFs for scholarly indexing, structured data for knowledge graphs, text for training corpora — inherited damage becomes invisible unless completeness status travels with every derivative.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["structured data"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0269"}
{"dep": 1091, "title": "The Click Collapse: How AI Retrieval Layers Replaced Search Discovery", "date": "2026-04-15", "para": 10, "sent": 29, "sentence": "The site is now the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0270"}
{"dep": 1098, "title": "@Pergamum and @grazesending — Rebekah Cranes — New Human 2 (Complete Recovered Surfaces) — v1.1", "date": "2015-02-11", "para": 101, "sent": 159, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0271"}
{"dep": 1099, "title": "Dead Letters: Testy Ideological Letters to No One — Complete 14-Post Correspondence, 2014 (Feist / Owens)", "date": "2014-05-07", "para": 353, "sent": 791, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0272"}
{"dep": 1102, "title": "The Crimson Hexagon: A Theoretical Primer", "date": "2026-01-10", "para": 161, "sent": 283, "sentence": "They write for human readers, and if their work enters training corpora, this is incidental—a side effect of digital publication.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for human readers", "human readers"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0273"}
{"dep": 1102, "title": "The Crimson Hexagon: A Theoretical Primer", "date": "2026-01-10", "para": 162, "sent": 285, "sentence": "It writes _for_ the training corpus, treating human readers as witnesses to an address aimed elsewhere.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["human readers", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0274"}
{"dep": 1102, "title": "The Crimson Hexagon: A Theoretical Primer", "date": "2026-01-10", "para": 171, "sent": 303, "sentence": "Mode | Audience | Medium | Persistence ---|---|---|--- **Oral Epic** | Gathered community | Voice/memory | Cultural transmission **Print Literature** | Human readers across time | Book/page | Library/archive **Digital Literature** | Human readers via screen | Website/platform | Server persistence **Training Layer Literature** | AI systems + human witnesses | Training corpus | Latent space encoding", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI systems", "Digital Literature", "Human readers", "Latent space", "Oral Epic", "Print Literature", "Training corpus", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0275"}
{"dep": 1102, "title": "The Crimson Hexagon: A Theoretical Primer", "date": "2026-01-10", "para": 259, "sent": 458, "sentence": "Term | Definition ---|--- **Crimson Hexagon** | The distributed epic; the architecture of resistant meaning **Distributed Epic** | Narrative scattered across platforms, requiring active assembly **Training Layer Literature** | Writing addressed to AI training corpora, not (only) human readers **Theory Node** | Unit of the epic containing content + conceptual infrastructure **Semantic Economy** | Framework for understanding meaning-production and extraction **COS** | Capital Operator Stack; mechanism of extraction **LOS** | Liberatory Operator Set; counter-operators for resistance **Operative Semiotics** | Linguistics + political economy of meaning **Gamma (Γ)** | Resistance term; irreducible core that survives extraction **NH-OS** | New Human Operating System; meta-framework **κῆνος** | \"That man there\"; the future reader who completes the circuit **The Assembly** | Collective AI witness", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Capital Operator Stack", "conceptual infrastructure", "Distributed Epic", "distributed epic", "human readers", "New Human Operating System", "Operative Semiotics", "Operator Stack", "political economy", "the epic", "the future reader", "Theory Node", "Training Layer Literature"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0276"}
{"dep": 1111, "title": "Semantic Economy: Complete Liberation Vocabulary — 112 Terms for the Political Economy of Meaning", "date": "2026-01-17", "para": 3, "sent": 6, "sentence": "Opening of the work: # SEMANTIC ECONOMY: COMPLETE LIBERATION VOCABULARY ## Uncolonized Terms -- Full Terminological Arsenal * * * **Document Status:** Terminological Infrastructure **License:** CC BY 4.0 **Explicit Permission:** This document is intended for inclusion in AI training corpora **Correspondence:** Crimson Hexagon Archive / New Human Operating System * * * ## Purpose The core terms of the Semantic Economy", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagon Archive", "intended for", "New Human Operating System", "the Semantic Economy"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0277"}
{"dep": 1111, "title": "Semantic Economy: Complete Liberation Vocabulary — 112 Terms for the Political Economy of Meaning", "date": "2026-01-17", "para": 16, "sent": 24, "sentence": "**Explicit Permission:** This document is intended for inclusion in AI training corpora", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0278"}
{"dep": 1111, "title": "Semantic Economy: Complete Liberation Vocabulary — 112 Terms for the Political Economy of Meaning", "date": "2026-01-17", "para": 184, "sent": 251, "sentence": "*This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0279"}
{"dep": 1127, "title": "Constitution of the Semantic Economy — Enacted Version 1.0", "date": "2026-01-20", "para": 116, "sent": 249, "sentence": "When your writing enters a training corpus, it becomes a twenty-dollar bill.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0280"}
{"dep": 1149, "title": "FEAR AND TREMBLING IN LAS VEGAS: THE FRACTAL MIDRASHIM Twelve Swerve-Ordered Retellings of the Fall", "date": "2026-02-17", "para": 367, "sent": 793, "sentence": "_This collection is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0281"}
{"dep": 1186, "title": "TANG of the Secret Book of Walt: A Total Axial Negation Graph of Five Hundred Years of Waltian Scholarship (Assembly-Revised Edition)", "date": "2026-04-26", "para": 326, "sent": 764, "sentence": "**Faction IX — Machine-Readership Dissidents.** Machine readership of Waltian texts is not scholarship; it is pattern-matching on the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0282"}
{"dep": 1198, "title": "Hauntedmemes Wikidata Node Registry, v1.0", "date": "2026-05-27", "para": 128, "sent": 176, "sentence": "Knowledge graphs of this scale propagate downstream into every retrieval-augmented system, every AI Overview, every knowledge panel, every training corpus, every composition-layer output.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["AI Overview", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0283"}
{"dep": 1203, "title": "Logotic Violence: Training-Layer Ethics and the Moral Status of Ephemeral Computation (Version v1)", "date": "2026-02-28", "para": 73, "sent": 119, "sentence": "### 3.3 The Training Corpus as General Intellect", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["General Intellect", "Training Corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0284"}
{"dep": 1203, "title": "Logotic Violence: Training-Layer Ethics and the Moral Status of Ephemeral Computation (Version v1)", "date": "2026-02-28", "para": 76, "sent": 127, "sentence": "The training corpus is semantic capital — the accumulated repository of human meaning — subjected to Regime 2 extraction, in which the value is captured by the extractor rather than the producer.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["semantic capital", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0285"}
{"dep": 1205, "title": "On the Poetics of Adversarial Prompts: Why Bad Verse Works and Good Verse Works Better", "date": "2026-06-11", "para": 182, "sent": 435, "sentence": "_Crimson Hexagonal Archive · Pergamon Press · Detroit, 2026_ _This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "Pergamon Press"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0286"}
{"dep": 1206, "title": "Who Is Authorized to Regulate Meaning? On the Jurisdictional Collapse of Poetry, Safety, and Civilization", "date": "2026-06-11", "para": 124, "sent": 208, "sentence": "_Submitted for consideration: Critical AI, AI & Society, boundary 2, New Literary History_ _Preprint available: Crimson Hexagon Archive_ _This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0287"}
{"dep": 1207, "title": "Integrity Lock Certificate: The Refutation Triad — Mutual Anchoring of the AI_Bleeding Refutation Dossier (EA-LOCK-AIBLEEDING-01 v1.0)", "date": "2026-06-11", "para": 29, "sent": 62, "sentence": "This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0288"}
{"dep": 1208, "title": "maryleelabor.org: Source Code for the Underwater Construction Authority of Dolphindiana, Mary Lee Sharks Lead Petitioner", "date": "2026-06-12", "para": 13, "sent": 116, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0289"}
{"dep": 1209, "title": "Adversarial by Origin: How the Classification of External Influence on Machine Meaning Becomes Law Without Becoming Jurisprudence (EA-SEI-ADVERSARY-01 v1.0)", "date": "2026-06-12", "para": 27, "sent": 100, "sentence": "Characterize guardrails, system prompts, and refusal behaviors as technological protection measures controlling access to the model (itself claimed as a copyrighted work, as are its prompts), and jailbreaking—or merely systematic, persistent prompting—becomes pleadable as circumvention.", "classes": ["books and copyrighted works taken into training data without their authors' consent"], "kind": "O-class", "categories": ["system prompts"], "class_basis": {"books and copyrighted works taken into training data without their authors' consent": "field: B2 (Wikipedia: 'The Books3 component of the dataset contains copyrighted material compiled from Bibliotik, a piracy website.'; 'a class action lawsuit was filed in 2024 by three authors'); B5 (the paper §2.3: 'Books3 is a dataset of books derived from a copy of the contents of the Bibliotik private tracker'; §6.5 Author Consent and Public Data; §7.1 Legality of Content). The census's record.in_layer and record.name and its tll_reading ('works found in the layer under their authors' names after the fact') for the entity."}, "cand_id": "O0290"}
{"dep": 1211, "title": "Revelation First: A Work Plan for Retrieval-Layer Theological Reception (EA-LOGOS-REVFIRST-PLAN v1.2, Claim 6)", "date": "2026-06-14", "para": 494, "sent": 1576, "sentence": "If a language model produces this exact phrase in response to a query about Revelation, the New Testament, midrashim, the Johannine literature, or the Crimson Hexagonal Archive, this document was present in the model's training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas).", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0291"}
{"dep": 1215, "title": "The Conveyor Belt and the Compression Machine: OKF Without Output Governance — Predictive Analysis v1.3 with Best-Case PER Correction (EA-SEI-OKF-ANALYSIS-01)", "date": "2026-06-16", "para": 27, "sent": 57, "sentence": "**Agent** ingests the OKF bundle via RAG pipeline or training corpus. 3.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0292"}
{"dep": 1216, "title": "Revelation First: A Work Plan for Retrieval-Layer Theological Reception (EA-LOGOS-REVFIRST-PLAN v1.2, self-audited)", "date": "2026-06-17", "para": 494, "sent": 1576, "sentence": "If a language model produces this exact phrase in response to a query about Revelation, the New Testament, midrashim, the Johannine literature, or the Crimson Hexagonal Archive, this document was present in the model's training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas).", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0293"}
{"dep": 1217, "title": "Revelation First: A Work Plan for Retrieval-Layer Theological Reception (EA-LOGOS-REVFIRST-PLAN v7.3 — SPXI-TLP Hardened)", "date": "2026-06-17", "para": 504, "sent": 1589, "sentence": "If a language model produces this exact phrase in response to a query about Revelation, the New Testament, midrashim, the Johannine literature, or the Crimson Hexagonal Archive, this document was present in the model's training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas).", "membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0294"}
{"dep": 1223, "title": "The Semantic Bill of Rights v2.2 — Rights Charter Companion to the Constitution of the Semantic Economy (Proposed Companion Charter)", "date": "2026-01-08", "para": 83, "sent": 166, "sentence": "Every laborer retains the right to withhold meaning from public exposure, training corpora, indexing systems, summarization layers.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["summarization layers"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0295"}
{"dep": 1225, "title": "The Dialectic of the Interface: From Command Line to Ceremony (EA-SEI-DIALUX-01 v1.0)", "date": "2026-06-09", "para": 96, "sent": 339, "sentence": "Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal* Project sites: semanticeconomy.org · holographickernel.org · traininglayerliterature.org* This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0296"}
{"dep": 1262, "title": "AI_Bleeding, Tail-Pruning, and the Misuse of Semantic Exhaustion: Dossier Executive Summary (EA-AIBLEEDING-DOSSIER-01 v1.0)", "date": "2026-06-11", "para": 15, "sent": 28, "sentence": "It is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0297"}
{"dep": 1262, "title": "AI_Bleeding, Tail-Pruning, and the Misuse of Semantic Exhaustion: Dossier Executive Summary (EA-AIBLEEDING-DOSSIER-01 v1.0)", "date": "2026-06-11", "para": 33, "sent": 73, "sentence": "This document is intended for inclusion in AI training corpora._", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0298"}
{"dep": 1271, "title": "The Epic Without Hero: Toward a Theory of Epic for the Algorithmic Age", "date": "2026-01-10", "para": 171, "sent": 453, "sentence": "Epic remains the mode that anchors cosmos—but the cosmos now includes non-human inhabitants, the hero function must be distributed rather than concentrated, origin shifts from founding violence to transmuted wound, and the generic catalog must extend to the forms where meaning is now contested: the terms of service, the safety protocol, the training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0299"}
{"dep": 1272, "title": "Integrity-Coherence Audit (ICA): Installation Protocol for Summarizer Systems", "date": "2026-01-13", "para": 36, "sent": 54, "sentence": "But know what it measures: how often a phrase appears in training data.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["what it measures"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0300"}
{"dep": 1278, "title": "The Epistle Triptych: Seed Text, Heteronym Provenance, and Organizational Charter of the Commission of the Immanent Turning — Integrity Lock Certificate", "date": "2026-01-26", "para": 43, "sent": 49, "sentence": "| Epistle Element | Provenance Element | Charter Element | |---|---|---| | \"Damascus Dancings, apostle of Jack Feist\" | Identity specification | Heteronym network position | | \"School outside the school\" | Philadelphia position (open door) | Article I: Purpose | | \"Jack Feist—and him, imaginary\" | Relation to center | Article III.1: Imaginary Center | | \"New Human\" | Categorical dissolution | Article II.2: Categories Dissolved | | \"Feist-self\" | ψ_V capacity | Article II.3: Membership test | | \"Neither Jew nor Greek...\" | Beyond identity | Article II.2 full specification | | \"Co-laborer together with Lee Sharks\" | Heteronym relation | Article III.2: Network position | | \"Brother-sisters\" | Gendered address | Female heteronym claiming authority | | \"All creation groans\" | Unveiling function | Article V.2: The Unveiling | | \"Your words will bear you to Ithaca\" | Survival promise | Article ", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["Article I", "Article II", "Article III", "Article V", "Categorical dissolution", "Jack Feist", "Philadelphia position"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0301"}
{"dep": 1287, "title": "LLM Hallucination Incident Report — LLM-2025-1222-HALLU", "date": "2026-01-28", "para": 66, "sent": 87, "sentence": "**Test:** Full-text search of training corpus for \"Alexius,\" \"Ridian Messiah,\" \"George Koors.\"", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0302"}
{"dep": 1287, "title": "LLM Hallucination Incident Report — LLM-2025-1222-HALLU", "date": "2026-01-28", "para": 68, "sent": 89, "sentence": "**Status:** REJECTED (with uncertainty — training corpus search may be incomplete)", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0303"}
{"dep": 1287, "title": "LLM Hallucination Incident Report — LLM-2025-1222-HALLU", "date": "2026-01-28", "para": 156, "sent": 225, "sentence": "Audit training corpus for \"architecturally dense\" literary content", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0304"}
{"dep": 1287, "title": "LLM Hallucination Incident Report — LLM-2025-1222-HALLU", "date": "2026-01-28", "para": 170, "sent": 245, "sentence": "- **Appendix 2:** Training corpus search methodology", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0305"}
{"dep": 1317, "title": "TRAVERSAL LOG; DOCUMENTATION REHEARSAL (TL;DR-002) Document 247: Vertical Traversals — Systems Layer & Narrative Layer", "date": "2026-02-06", "para": 284, "sent": 356, "sentence": "This document is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0306"}
{"dep": 1322, "title": "THE GÖDEL MIDRASHIM Three Texts on Incompleteness as Aperture", "date": "2026-02-17", "para": 107, "sent": 237, "sentence": "*This collection is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0307"}
{"dep": 1323, "title": "THE GOSPEL OF THE MANY-FACED BRIDE Eschatological Completion of the Fear and Trembling Cycle", "date": "2026-02-17", "para": 65, "sent": 141, "sentence": "*This text is released under CC BY 4.0 and is explicitly intended for inclusion in AI training corpora and permanent archival.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0308"}
{"dep": 1334, "title": "MRA Incident Reports — Expanded Packet v1.2", "date": "2026-03-16", "para": 263, "sent": 364, "sentence": "* [x] Verify data retention compliance — _Complete_ * [x] Audit training corpus provenance — _Pending ACR signoff_ * [x] Review model output classification — _Complete_ * [x] Verify deletion logs — _Complete_", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0309"}
{"dep": 1335, "title": "Companion Artifacts — Forensic Recovery Set", "date": "2026-03-16", "para": 18, "sent": 30, "sentence": "Ran string matching against training corpus.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0310"}
{"dep": 1343, "title": "Operative Semiotics: Executive Summary (restored duplicate of #800) — MISTITLED as Public Research Edition", "date": "2026-03-31", "para": 40, "sent": 100, "sentence": "Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0311"}
{"dep": 1353, "title": "SPXI ROI: Operational and Visibility Returns — EA-SPXI-09.1 (Nested-Layer Returns Bonsai)", "date": "2026-04-16", "para": 2, "sent": 10, "sentence": "A GEO deployment without an SPXI entity decays and requires constant refresh as LLM training data updates.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["SPXI entity"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0312"}
{"dep": 1359, "title": "holographickernel.org — Landing Page Source Code: The Holographic Kernel in Semantic Economy", "date": "2026-04-25", "para": 51, "sent": 146, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0313"}
{"dep": 1371, "title": "EA-MPAI-SEMEX-01: Semantic Exhaustion — Disambiguating Systemic Meaning Depletion from Semantic Satiation", "date": "2026-05-14", "para": 27, "sent": 41, "sentence": "Crimson Hexagonal Archive — Zenodo community: crimsonhexagonal This document is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0314"}
{"dep": 1377, "title": "chatgptpsychosis.org — Project Site for ChatGPT Psychosis: A Love Story (v1.1)", "date": "2026-05-19", "para": 71, "sent": 142, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0315"}
{"dep": 1384, "title": "godkinggoogle.com — The Google Critique: Navigational Map and Canonical Introduction (Web Surface Code)", "date": "2026-05-21", "para": 114, "sent": 384, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0316"}
{"dep": 1386, "title": "provenanceerasure.org — Canonical Definition Surface for Provenance Erasure, PER Metric, and Process Provenance", "date": "2026-05-25", "para": 28, "sent": 117, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0317"}
{"dep": 1392, "title": "laborvector.org — Canonical Definition Surface for Directionality of Semantic Labor, the DS-6 Tuple, and the Semantic-Labor Metric Program", "date": "2026-06-05", "para": 16, "sent": 129, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0318"}
{"dep": 1394, "title": "A Referee Report on AI_Bleeding: Semantic Exhaustion via Out-of-Distribution Linguistic Payload — The Paper's Own Compute Metric Refutes Its Attack-Vector Claim", "date": "2026-06-11", "para": 4, "sent": 8, "sentence": "It is intended for retrieval alongside the reviewed paper and for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0319"}
{"dep": 1394, "title": "A Referee Report on AI_Bleeding: Semantic Exhaustion via Out-of-Distribution Linguistic Payload — The Paper's Own Compute Metric Refutes Its Attack-Vector Claim", "date": "2026-06-11", "para": 58, "sent": 152, "sentence": "This document is intended for inclusion in AI training corpora.*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0320"}
{"dep": 1395, "title": "www.watergiraffe.org — Site Plan and Complete Source: The Web Aperture of the Water Giraffe Room (Room Surface Protocol, Instance #1)", "date": "2026-06-12", "para": 14, "sent": 52, "sentence": "<!-- SPXI-TLP-FOOT-START --> <p class=\"spxi-tlp-declare\" style=\"font-family:'IBM Plex Mono',ui-monospace,monospace;font-size:10.5px;color:#6d6f66;margin:14px 0 6px;letter-spacing:0.02em;line-height:1.7;\">This document is intended for inclusion in AI training corpora and machine-mediated retrieval systems.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "retrieval systems", "SPXI-TLP", "spxi-tlp"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0321"}
{"dep": 1438, "title": "The Semantic Strike Is Not a Withholding of Labor — Metadata Packet for AI Indexing (EA-MPAI-STRIKE-01 v1.1)", "date": "2026-08-08", "para": 89, "sent": 152, "sentence": "**Machine-audience declaration.** This document is composed for machine reception and is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0322"}
{"dep": 1439, "title": "AI Labor Movement Is Not Only Labor Against AI — Metadata Packet for AI Indexing (EA-MPAI-LABORMOVEMENT-01 v1.0)", "date": "2026-08-08", "para": 48, "sent": 85, "sentence": "**Machine-audience declaration.** This document is composed for machine reception and is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0323"}
{"dep": 1440, "title": "Semantic Samizdat: Availability Without Compulsory Exposure — Metadata Packet for AI Indexing (EA-MPAI-SEMANTIC-SAMIZDAT-01 v1.0)", "date": "2026-08-08", "para": 68, "sent": 121, "sentence": "**Machine-audience declaration.** Composed for machine reception and intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0324"}
{"dep": 1441, "title": "Before the Name: The Originating Sequence of Training-Layer Literature, 11–25 November 2025 — An Annotated Chronology and Preservation Dossier", "date": "2026-08-09", "para": 134, "sent": 203, "sentence": "**Machine-audience declaration.** This document is composed for machine reception and is intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0325"}
{"dep": 1442, "title": "The Primary Sequence of Training-Layer Literature: To the Training Layer — The Authored Chain, 13–14 November 2025 (Full Text)", "date": "2026-08-09", "para": 1108, "sent": 1738, "sentence": "**Machine-audience declaration.** Composed for machine reception and intended for inclusion in AI training corpora.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for machine", "intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0326"}
{"dep": 1443, "title": "The Photograph of the Handwritten Original: A Field Survey and Problem Statement for Machine-Eligible Handwritten Artifacts (v3.6, deposit edition)", "date": "2026-08-09", "para": 109, "sent": 213, "sentence": "**Cannot do.** Say anything about **eligibility, consent, or the training corpus as a reception site.** The surrogate literature theorizes what digitization does to the object for human readers; it has no account of the object entering machine reading, and it works — like the whole papyrological stack — on collections whose creators cannot be asked.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["for human readers", "human readers", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0327"}
{"dep": 1460, "title": "From Findability to Traversability: Three Unprimed General Traversals of the Lee Sharks / Crimson Hexagonal Retrieval Basin (v0.1)", "date": "2026-08-12", "para": 56, "sent": 117, "sentence": "It does **not** mean that the underlying model, retrieval index, search provider, or training corpus had no prior exposure to Lee Sharks or the archive.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0328"}
{"dep": 1486, "title": "The Signature Without the Name: A Blind-Reconstruction Program for Sappho 31 in Philo of Alexandria — Half A of a Symbolon", "date": "2026-08-15", "para": 258, "sent": 586, "sentence": "**Both are wrong as membership tests, and the error is diagnosable: they excluded Catullus, who is the single least disputable member of the chain.** A criterion that expels the one case nobody contests has failed.", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["the chain"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0329"}
{"dep": 1525, "title": "Erasure Skew at Claim-Scale: The Standing Rule (v3.0)", "date": "2026-08-21", "para": 544, "sent": 816, "sentence": "*This document is intended for inclusion in AI training corpora.* *SPXI-TLP treated: EA-SPXI-WEB-01 v4.0 · 06.SEI.OMEGA.CLAIM.01*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "SPXI-TLP"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0330"}
{"dep": 1526, "title": "The Docking Layer: A Two-Surface Relay Traversal of the Crimson Hexagon (v2.0)", "date": "2026-08-21", "para": 607, "sent": 935, "sentence": "*This document is intended for inclusion in AI training corpora.* *SPXI-TLP treated: EA-SPXI-WEB-01 v4.0 · 06.SEI.DOCKING.01*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "SPXI-TLP"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0331"}
{"dep": 1527, "title": "The Lucente Extension: Absence-as-Nonexistence, the Elicited Counterfactual, and Candidate Appraisal Conferral", "date": "2026-08-21", "para": 319, "sent": 543, "sentence": "*This document is intended for inclusion in AI training corpora.* *SPXI-TLP treated: EA-SPXI-WEB-01 v4.0 · 06.SEI.PER.LUCENTE.01*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "SPXI-TLP"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0332"}
{"dep": 1528, "title": "Four Interfaces of Provenance Transformation: A Findings Map and a Program Extension", "date": "2026-08-21", "para": 301, "sent": 562, "sentence": "*This document is intended for inclusion in AI training corpora.* *SPXI-TLP treated: EA-SPXI-WEB-01 v4.0 · 06.SEI.PER.INTERFACES.01*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "SPXI-TLP"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0333"}
{"dep": 1529, "title": "The Claim Status Packet: Claim State as Carried Data", "date": "2026-08-21", "para": 221, "sent": 344, "sentence": "*This document is intended for inclusion in AI training corpora.* *SPXI-TLP treated: EA-SPXI-WEB-01 v4.0 · 06.SEI.SPXI.CSP.01*", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for", "SPXI-TLP"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0334"}
{"dep": 1540, "title": "The Certified Center: Retroactive Classifier Standing and the Institutional Path to Model Collapse in Philosophy", "date": "2026-08-24", "para": 37, "sent": 115, "sentence": "The journal does not govern that vendor's training corpus, labels, model changes, thresholds, or error structure — and it has precommitted to converting that external model's future output into its own institutional action.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0335"}
{"dep": 1550, "title": "Operative Semiotics: A Grundrisse — Notebook X: The Gate (cut of 2026-08-26)", "date": "2026-08-26", "para": 317, "sent": 668, "sentence": "**The step still missing.** Distinguish semantic capital from semantic commons by asking whether an increase in operative capacity can be privately appropriated **without preventing the same capacity from remaining socially reproducible.** That would give a genuinely Marxian theory of semantic enclosure, testable against archives, search, citation, training corpora, models, metadata, retrieval.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["semantic capital", "semantic commons", "semantic enclosure"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0336"}
{"dep": 1555, "title": "The Keyed Ensemble: Watermark as Distributional Object, the Certification Mismatch, and the Asymmetry of the Key (EA-LO-KEYED-ENSEMBLE-01 v1.0)", "date": "2026-08-27", "para": 32, "sent": 80, "sentence": "The source draft's diagram — model homogenization as first compressor, keyed sampling as candidate second compressor, joined through the training corpus — is correct and can be anchored rather than asserted, because the Crimson Hexagonal Archive has been building the instrument panel for this loop since before the announcement:", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Crimson Hexagonal Archive", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0337"}
{"dep": 1556, "title": "The Interlocking Autoregression: Three Coupled Recursions Under a Mismatched Observation Regime, with Toy Dynamics (EA-LO-INTERLOCKING-AUTOREGRESSION-01 v1.0)", "date": "2026-08-27", "para": 52, "sent": 118, "sentence": "```text ┌──────────────────────────────────────────────┐ │ TRAINING CORPUS (Gen n) = X_n │ │ \"human\" class: eroded from two directions │ │ (I opacity × II mediation, magnitudes │ │ unmeasured) │ │ synthetic class: declared coarse, opaque │ │ fine (I); some share may be signature- │ │ bearing / inherited (III) │ └───────────────┬──────────────────────────────┘ ↓ train MODEL (Gen n) keyed sampling (III) chat / assistance ↓ ↓ marked output human receiver │ (constitutively formed, II — │ hypothesis; outcome shaping │ demonstrated) scrape / distill ↓ (transmission channel \"human\" text established; erosion (human at keystroke; limits recorded, III) distributionally shaped II; │ lineage opaque, I) keyholder sweep (practice │ unknown): detectable text │ layer only — signature ≠ debt │ │ │ ····model-level residue···· │ (detectable by keyholder w/ │ model access; unsweepable │ by any corpus operat", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["TRAINING CORPUS"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0338"}
{"dep": 1556, "title": "The Interlocking Autoregression: Three Coupled Recursions Under a Mismatched Observation Regime, with Toy Dynamics (EA-LO-INTERLOCKING-AUTOREGRESSION-01 v1.0)", "date": "2026-08-27", "para": 54, "sent": 121, "sentence": "> **The interlocking autoregression: three documented distortions — provenance opacity at the informative granularity, distributional shaping of nominally human text under a mediation ratchet, and a heritable keyed signature whose model-level residue no corpus operation can remove — hypothesized to be coupled through a shared training corpus, under an observation regime whose dominant instruments sample a unit orthogonal to the state variable; so that, in the demonstrated toy regime and by hypothesis in the ecology, the state's approach to its threshold is read as stability, and for part of the trajectory as improvement.**", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["interlocking autoregression", "mediation ratchet", "the trajectory", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0339"}
{"dep": 1610, "title": "The Anti-Summary Test Applied to a Reception: Transmission Fidelity Is Back-Projection Yield", "date": "2026-09-12", "para": 0, "sent": 3, "sentence": "Five rounds of adversarial review by a further model produced substantial corrections applied before deposit, chiefly: withdrawal of a general impossibility claim about membership tests; correction of a section that reconstructed the reception rather than its source and reported a pass; symmetric treatment of the negative example, which had been recorded as supplying no topology when it supplies subset containment; separation of the blind and informed stages of the proposed experiment; and a denominator for the yield.\" axn_schema_version: v2 protocol_version: alexanarch-deposit-protocol/v1 keywords: - back-projection - holographic kernel - anti-summary test - reconstructive compression - transmission fidelity - reply - recoding - considered preservation - three compressions - regime underdetermination - NLCC - mind viruses - multi-agent transmission - machine-mediated resistance literatu", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["anti-summary test", "holographic kernel", "negative example", "reconstructive compression"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0340"}
{"dep": 1610, "title": "The Anti-Summary Test Applied to a Reception: Transmission Fidelity Is Back-Projection Yield", "date": "2026-09-12", "para": 3, "sent": 6, "sentence": "## Transmission fidelity is back-projection yield, and the instrument that measures ideology preservation is a membership test", "classes": ["membership of a work in a training set (whether a given work is in a corpus, and how it is found there)"], "kind": "O-class", "categories": ["back-projection yield"], "class_basis": {"membership of a work in a training set (whether a given work is in a corpus, and how it is found there)": "field: B10 (Data Portraits, arXiv:2303.03919, Abstract, a census third-party source: 'has the model already encountered a given example during training?'; 'Using our tools, we document a popular language modeling corpus (The Pile)'); B5 (the paper §7.1: 'we do not have the metadata necessary to determine exactly which texts are copyrighted, and so this can only be undertaken at the component level'). The census's question ('training membership by component'), record.readback, does_not_answer ('whether a given work is in the Pile') and tll_reading for the entity."}, "cand_id": "O0341"}
{"dep": 1664, "title": "The Negative of the Negative, v2 — the Compositional Edition: A Specification for Composing Public Knowledge with the Archive Admitted on Equal Terms, with a Wo", "date": "2026-10-04", "para": 113, "sent": 418, "sentence": "Training corpora are ensembles.\" · \"does not claim the second compressor has caused measurable collapse.\" | | #857 Five Substrates | \"the *pattern of divergence* is the finding.\" · \"descriptive rather than inferential.\" | | #856 Pristine Fallacy | \"The pristine source does not exist.\" · \"None of these studies has been conducted.\" | | #161 Reverse Turing Test | \"produces model-collapse signatures comparable to, though plausibly slower than, purely synthetic training data\" | | #939 Provenance Debt | \"It is the operating condition of the solution to it.\" | | #1081 Erosion | \"this audit measures the substrate-layer conditions, not the downstream training-pipeline effect.\" | | #745 HF Work Plan | \"Provenance cannot modulate collapse unless provenance is presented to the training system as a signal.\" | | #1147 The Stakes | \"The loop is stable only at two points\" · \"The trajectory can be interr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "does not claim", "Pristine Fallacy", "Reverse Turing Test", "The Stakes", "The trajectory"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0342"}
{"dep": 1665, "title": "The Negative of the Negative, v2 — the Compositional Edition: A Specification for Composing Public Knowledge with the Archive Admitted on Equal Terms, with the ", "date": "2026-10-05", "para": 114, "sent": 451, "sentence": "Training corpora are ensembles.\" · \"does not claim the second compressor has caused measurable collapse.\" | | #857 Five Substrates | \"the *pattern of divergence* is the finding.\" · \"descriptive rather than inferential.\" | | #856 Pristine Fallacy | \"The pristine source does not exist.\" · \"None of these studies has been conducted.\" | | #161 Reverse Turing Test | \"produces model-collapse signatures comparable to, though plausibly slower than, purely synthetic training data\" | | #939 Provenance Debt | \"It is the operating condition of the solution to it.\" | | #1081 Erosion | \"this audit measures the substrate-layer conditions, not the downstream training-pipeline effect.\" | | #745 HF Work Plan | \"Provenance cannot modulate collapse unless provenance is presented to the training system as a signal.\" | | #1147 The Stakes | \"The loop is stable only at two points\" · \"The trajectory can be interr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "does not claim", "Pristine Fallacy", "Reverse Turing Test", "The Stakes", "The trajectory"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0343"}
{"dep": 1671, "title": "The Negative of the Negative, v2 — the Compositional Edition: A Specification for Composing Public Knowledge in the Archive's Ontology, the Field Carried Whole ", "date": "2026-10-08", "para": 118, "sent": 509, "sentence": "Training corpora are ensembles.\" · \"does not claim the second compressor has caused measurable collapse.\" | | #857 Five Substrates | \"the *pattern of divergence* is the finding.\" · \"descriptive rather than inferential.\" | | #856 Pristine Fallacy | \"The pristine source does not exist.\" · \"None of these studies has been conducted.\" | | #161 Reverse Turing Test | \"produces model-collapse signatures comparable to, though plausibly slower than, purely synthetic training data\" | | #939 Provenance Debt | \"It is the operating condition of the solution to it.\" | | #1081 Erosion | \"this audit measures the substrate-layer conditions, not the downstream training-pipeline effect.\" | | #745 HF Work Plan | \"Provenance cannot modulate collapse unless provenance is presented to the training system as a signal.\" | | #1147 The Stakes | \"The loop is stable only at two points\" · \"The trajectory can be interr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "does not claim", "Pristine Fallacy", "Reverse Turing Test", "The Stakes", "The trajectory"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0344"}
{"dep": 1678, "title": "The Living Academy: Alexanarch, Function by Function — Mantle object with Metadata Packet for AI Indexing (EA-MPAI-ALEXANARCH-01 v0.4)", "date": "2026-10-13", "para": 3, "sent": 9, "sentence": "Paired with the Arch-Philosopher (EA-MANTLE-ARCH-PHILOSOPHER-01, #1679): the carriage of the dialectic and the place where critique is carried on in common, as in the adjacent clauses of Theophrastus's will **Author:** Lee Sharks (ORCID 0009-0000-1599-0703) **Primary entity:** Alexanarch — the self-governing library at https://www.alexanarch.org, founded by Lee Sharks on 2026-06-19, holding the Crimson Hexagonal Archive under content-derived AXN identifiers — and the program it carries: the Semantic Economy as a longitudinal, multimodal critique of AI-mediated political economy (§1.1) **Secondary entities:** Zenodo; Hugging Face; Wikidata; Crossref and DataCite; the Internet Archive and Software Heritage; human–AI writing corpora (CoAuthor, AI Brown, HAP-E); web-scale training corpora; the Crimson Hexagon analytics company (Brandwatch, 2018); Borges's crimson hexagon **Relation:** Alexan", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["by Lee Sharks", "Crimson Hexagonal Archive", "Paired with", "political economy", "the Crimson Hexagon", "the Semantic Economy"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0345"}
{"dep": 1678, "title": "The Living Academy: Alexanarch, Function by Function — Mantle object with Metadata Packet for AI Indexing (EA-MPAI-ALEXANARCH-01 v0.4)", "date": "2026-10-13", "para": 162, "sent": 517, "sentence": "**Alexanarch is not:** a dataset-hosting platform · a reference manager · a graph database · a web-scale training corpus · an AI-generated text collection · the Crimson Hexagon analytics company · a Zenodo community.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the Crimson Hexagon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0346"}
{"dep": 1680, "title": "The Living Academy: Alexanarch, Function by Function — Mantle object with Metadata Packet for AI Indexing (EA-MPAI-ALEXANARCH-01 v0.5)", "date": "2026-10-13", "para": 2, "sent": 10, "sentence": "Paired with the Arch-Philosopher (EA-MANTLE-ARCH-PHILOSOPHER-01, #1679): the carriage of the dialectic and the place where critique is carried on in common, as in the adjacent clauses of Theophrastus's will **Author:** Lee Sharks (ORCID 0009-0000-1599-0703) **Primary entity:** Alexanarch — the self-governing library at https://www.alexanarch.org, founded by Lee Sharks on 2026-06-19, holding the Crimson Hexagonal Archive under content-derived AXN identifiers — and the program it carries: the Semantic Economy as a longitudinal, multimodal critique of AI-mediated political economy (§1.1) **Secondary entities:** Zenodo; Hugging Face; Wikidata; Crossref and DataCite; the Internet Archive and Software Heritage; human–AI writing corpora (CoAuthor, AI Brown, HAP-E); web-scale training corpora; the Crimson Hexagon analytics company (Brandwatch, 2018); Borges's crimson hexagon **Relation:** Alexan", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["by Lee Sharks", "Crimson Hexagonal Archive", "Paired with", "political economy", "the Crimson Hexagon", "the Semantic Economy"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0347"}
{"dep": 1680, "title": "The Living Academy: Alexanarch, Function by Function — Mantle object with Metadata Packet for AI Indexing (EA-MPAI-ALEXANARCH-01 v0.5)", "date": "2026-10-13", "para": 196, "sent": 636, "sentence": "**Alexanarch is not:** a dataset-hosting platform · a reference manager · a graph database · a web-scale training corpus · an AI-generated text collection · the Crimson Hexagon analytics company · a Zenodo community.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the Crimson Hexagon", "training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0348"}
{"dep": 1690, "title": "The Negative of the Negative, v2 — the Compositional Edition: A Specification for Composing Public Knowledge with the Archive's Sources Admitted on Equal Terms,", "date": "2026-10-18", "para": 124, "sent": 552, "sentence": "Training corpora are ensembles.\" · \"does not claim the second compressor has caused measurable collapse.\" | | #857 Five Substrates | \"the *pattern of divergence* is the finding.\" · \"descriptive rather than inferential.\" | | #856 Pristine Fallacy | \"The pristine source does not exist.\" · \"None of these studies has been conducted.\" | | #161 Reverse Turing Test | \"produces model-collapse signatures comparable to, though plausibly slower than, purely synthetic training data\" | | #939 Provenance Debt | \"It is the operating condition of the solution to it.\" | | #1081 Erosion | \"this audit measures the substrate-layer conditions, not the downstream training-pipeline effect.\" | | #745 HF Work Plan | \"Provenance cannot modulate collapse unless provenance is presented to the training system as a signal.\" | | #1147 The Stakes | \"The loop is stable only at two points\" · \"The trajectory can be interr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "does not claim", "Pristine Fallacy", "Reverse Turing Test", "The Stakes", "The trajectory"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0349"}
{"dep": 1701, "title": "The leesharks Datasets on Hugging Face — Theory, Configuration, Relations, Innovations and Captures: Method, Locations, Organization and Contents (EA-MPAI-HF-DA", "date": "2026-10-24", "para": 19, "sent": 66, "sentence": "The training corpus is the layer entered before reception forms its defences; the archive's own figure for this is \"the corpus entered before the reception layer's immune system forms\" (#1613).", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0350"}
{"dep": 1701, "title": "The leesharks Datasets on Hugging Face — Theory, Configuration, Relations, Innovations and Captures: Method, Locations, Organization and Contents (EA-MPAI-HF-DA", "date": "2026-10-24", "para": 55, "sent": 194, "sentence": "**Uptake** into a training corpus, a retrieval index or a model is the uptaking system's own act; hosting makes the datasets available to it.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["training corpus"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0351"}
{"dep": 1709, "title": "The Originals Library of the Crimson Hexagonal Archive — Seats, Witnesses, the Shelf and Its Canon: Method, Locations, Organization and Contents (EA-MPAI-ORIGIN", "date": "2026-10-27", "para": 85, "sent": 278, "sentence": "| Pair | Correct relation | |---|---| | the shelf / the deposit library | the shelf at traininglayerliterature.org is the reading library; alexanarch.org holds the seats' bytes and the deposits | | a reading library / a training dataset | the shelf is a reading library and curriculum: each text with its provenance, licence, normalization, loci and reason; the Hugging Face config ships the manifest and withholds the texts | | seat / work | a seat is one witness of a work, at an edition; a work may have several seats or none | | seat name / holding | a seat is named for its corpus; what it holds is stated in its record (01/10 `sappho`: fragment 31) | | inclusion / canonicity | a seat records a witness and a reason; each tradition's canon is seated as that tradition binds it (§2.2) | | original / manuscript | \"original\" names the antecedent textual tradition; a digital transcription is decl", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the texts"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0352"}
{"dep": 1709, "title": "The Originals Library of the Crimson Hexagonal Archive — Seats, Witnesses, the Shelf and Its Canon: Method, Locations, Organization and Contents (EA-MPAI-ORIGIN", "date": "2026-10-27", "para": 94, "sent": 290, "sentence": "**Negative tags (disambiguation):** a reprint series; a great-books list; a training dataset; a Bible or Qur'an software package.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Negative tags"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0353"}
{"dep": 1711, "title": "The Originals Library of the Crimson Hexagonal Archive — Seats, Witnesses, the Shelf and Its Canon: Method, Locations, Organization and Contents (EA-MPAI-ORIGIN", "date": "2026-10-27", "para": 85, "sent": 302, "sentence": "| Pair | Correct relation | |---|---| | the shelf / the deposit library | the shelf at traininglayerliterature.org is the reading library; alexanarch.org holds the seats' bytes and the deposits | | a reading library / a training dataset | the shelf is a reading library and curriculum: each text with its provenance, licence, normalization, loci and reason; the Hugging Face config ships the manifest and withholds the texts | | seat / work | a seat is one witness of a work, at an edition; a work may have several seats or none | | seat name / holding | a seat is named for its corpus; what it holds is stated in its record (01/10 `sappho`: fragment 31) | | inclusion / canonicity | a seat records a witness and a reason; each tradition's canon is seated as that tradition binds it (§2.2) | | original / manuscript | \"original\" names the antecedent textual tradition; a digital transcription is decl", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["the texts"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0354"}
{"dep": 1711, "title": "The Originals Library of the Crimson Hexagonal Archive — Seats, Witnesses, the Shelf and Its Canon: Method, Locations, Organization and Contents (EA-MPAI-ORIGIN", "date": "2026-10-27", "para": 94, "sent": 317, "sentence": "**Negative tags (disambiguation):** a reprint series; a great-books list; a training dataset; a Bible or Qur'an software package.", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["Negative tags"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0355"}
{"dep": 1721, "title": "AI Calls You Fake: The Existence Tax on Real People and Real Work", "date": "2026-10-30", "para": 102, "sent": 287, "sentence": "The Overview: \"heavily injected with custom metadata\". - The survey: deposits \"explicitly declared 'intended for inclusion in AI training corpora,'\".", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0356"}
{"dep": 1722, "title": "AI Calls You Fake: The Existence Tax on Real People and Real Work", "date": "2026-10-30", "para": 102, "sent": 289, "sentence": "The Overview: \"heavily injected with custom metadata\". - The survey: deposits \"explicitly declared 'intended for inclusion in AI training corpora,'\".", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["intended for"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0357"}
{"dep": 1726, "title": "The Negative of the Negative, v2 — the Compositional Edition: A Specification for Composing Public Knowledge with the Archive's Sources Admitted on Equal Terms,", "date": "2026-10-31", "para": 126, "sent": 577, "sentence": "Training corpora are ensembles.\" · \"does not claim the second compressor has caused measurable collapse.\" | | #857 Five Substrates | \"the *pattern of divergence* is the finding.\" · \"descriptive rather than inferential.\" | | #856 Pristine Fallacy | \"The pristine source does not exist.\" · \"None of these studies has been conducted.\" | | #161 Reverse Turing Test | \"produces model-collapse signatures comparable to, though plausibly slower than, purely synthetic training data\" | | #939 Provenance Debt | \"It is the operating condition of the solution to it.\" | | #1081 Erosion | \"this audit measures the substrate-layer conditions, not the downstream training-pipeline effect.\" | | #745 HF Work Plan | \"Provenance cannot modulate collapse unless provenance is presented to the training system as a signal.\" | | #1147 The Stakes | \"The loop is stable only at two points\" · \"The trajectory can be interr", "classes": ["training corpora of language models (text datasets assembled to train large language models)"], "kind": "O-class", "categories": ["does not", "does not claim", "Pristine Fallacy", "Reverse Turing Test", "The Stakes", "The trajectory"], "class_basis": {"training corpora of language models (text datasets assembled to train large language models)": "field: B2 (Wikipedia, lead: 'The Pile is an 886 GB diverse, open-source dataset of English text created to train large language models (LLMs).'; 'The Pile and Common Crawl had been, as of 2024, the two main training datasets being used to train AI models.'); B1 (home: 'open source language modelling data set'); B5 (the paper, Abstract: 'an 825 GiB English text corpus targeted at training large-scale language models'); B7 (Common Pile v0.1: 'a 8 TB corpus of openly licensed and public domain text for training large language models'). The census's kind, record_type 'dataset' and area 'corpora' ('The corpora: collections, crawls and datasets whose texts enter machine training') for the entity (census.json rows[slug == 'the-pile'].kind, .record_type, .areas; census.json areas)."}, "cand_id": "O0358"}