{
 "spxi_instrument": {
  "method": "0.9",
  "name": "Capture Registry (longitudinal observation)",
  "level": 0,
  "level_name": "Measurement",
  "url": "https://www.spxi.dev/methods/#m-0-9",
  "inventory": "https://www.spxi.dev/methods/",
  "statement": "SPXI instrument: method 0.9 of 85, level 0 (Measurement), Capture Registry (longitudinal observation).",
  "also": []
 },
 "slug": "inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016",
 "date": "2026-10-16",
 "surface": "ChatGPT",
 "surface_basis": "The paste is a chatgpt.com conversation ('ChatGPT said:', 'Chat with ChatGPT', 'Ask ChatGPT').",
 "surfaces": [
  "ChatGPT"
 ],
 "auth": "signed out, incognito",
 "auth_basis": "Signed out: the paste carries the 'Log in' button. Incognito: operator default of 2026-10-14 ('theyre all signed out and incognito').",
 "ev": "paste",
 "mt": "ELEVEN OF FIFTEEN — THE WORKS ADDRESSED TO THE SUMMARIZER LEFT OUT",
 "match": "BROAD MATCH",
 "match_note": null,
 "s": "Archive",
 "q": "Inventory the leesharks datasets on hf, reading them as literature",
 "q_kind": "a reader's request to inventory the account's datasets as literature. NEW address. Asked the same day the operator ruled the Hub card should carry the account's datasets ('we should add them in the crimson-hexagonal description').",
 "cites": 1,
 "cite_list": [
  {
   "n": 1,
   "site": "Datasets at Hugging Face",
   "rel": "authored_surface",
   "title": null,
   "snip": null,
   "url": "https://huggingface.co/leesharks",
   "note": "chip shown 16 times ('+1' ×6, '+10' once); the leesharks dataset cards and their datasets-server views"
  }
 ],
 "archive_controlled_cites": 0,
 "transcript": "[ChatGPT (chatgpt.com), signed out (the paste carries 'Log in'), incognito (operator default of 2026-10-14). One turn; the paste's 'You said:' line is empty — the query is the operator's, given in his message of 2026-10-16 14:14 EDT. Source chips inline, site labels only ('Datasets at Hugging Face' ×16); image captions from third-party pages inline beside the dataset headings, kept as pasted.]\n\n[QUERENT] Inventory the leesharks datasets on hf, reading them as literature\n\n[ANSWER]\n\nThe leesharks datasets on Hugging Face: a literary inventory\nThe most revealing way to read the leesharks datasets on Hugging Face is not as a collection of unrelated machine-learning resources, but as a literary archive that has learned to represent itself as data.\n\nThe central work is the Crimson Hexagonal Archive, a corpus of authored texts, attributed personae, concepts, citations, revisions, and records of machine-mediated reception. Its Hugging Face representation makes these relationships queryable: the main dataset describes 1,625 deposits in the version documented by its dataset card, alongside separate tables for citations, terminology, and other archival functions. \n\nDatasets at Hugging Face\n+1\n\nThat changes the reading experience. A dataset row can function simultaneously as a catalogue entry, a miniature critical argument, a cross-reference, and a claim about what counts as evidence.\n\nI'll distinguish three things throughout this inventory:\n\nThe work: what a dataset contains and what literary or scholarly project it serves.\n\nThe form: how its tables, identifiers, classifications, and links shape what can be read.\n\nThe poetics: what the dataset implies about authorship, memory, authority, and the survival of meaning.\n\nThe inventory below is a researched map of the identifiable public datasets, not yet a guarantee that every repository under the account has been enumerated.\n\n1. The primary corpus: the archive as a book that can be queried\n\"Background abstract art red black texture illustration design\" Journal for Sale by MarketVisionAV | Redbubble\n1. crimson-hexagonal-archive\nThe primary corpus · 23 configurations\n\nOpen dataset ↗ \n\nThe archive's machine-readable edition: deposits, full texts, citations, coined vocabulary, reception records, predictions, and withdrawn or severed records. The documented build contains 1,625 deposits. Its 23 configurations are different views into one dataset, not 23 independent corpora. \n\nDatasets at Hugging Face\n+1\n\nLiterary reading: a book whose index, concordance, bibliography, editorial history, and apparatus have become separately addressable objects. Its most interesting formal feature is that relationships between texts are explicitly recorded rather than left entirely to the reader to reconstruct.\n\nCataloghi di libri e periodici | Archiginnasio\n2. heteronyms\nAuthorship as a plural system · 493 nodes\n\nOpen dataset ↗ \n\nMaps the Dodecad, other declared identities, collaborators, personae, and the machinery that attributes works to names. Its own documentation distinguishes twenty-six declared identities while noting that only twelve are fully covered by the apparatus. \n\nDatasets at Hugging Face\n\nLiterary reading: this is an anti-biographical work. Instead of asking which stable person stands behind each text, it makes the attribution system itself part of the text's meaning. The incomplete coverage matters: plurality is not merely a cast list but a problem of representation.\n\nProcesso: Linguagem e Comunicação | Wlademir Dias-Pino\n3. poetics\nTheory of literary making · 99 nodes\n\nOpen dataset ↗ \n\nA traversal through the archive's concepts of authorship, literary production, and related questions. The published view identifies 99 nodes. \n\nDatasets at Hugging Face\n\nLiterary reading: the closest thing here to an ars poetica, except that its argument is distributed across a graph. It asks us to read poetics as a set of relations and operations, not just as a statement of principles.\n\nReading the primary corpus\nThe main archive should be read before its smaller projections. Three features deserve particular attention:\n\nThe canonical text. The deposits configuration includes full text, title, attributed creator, date, description, content type, status, license, and substrate disclosure. This is where the literary object can be read rather than merely classified.\n\nThe citation graph. The citations configuration records directed relations between deposits. Reading becomes partly a matter of tracing how one text recruits, transforms, or contests another.\n\nThe archive's editorial time. Persistent identifiers, status fields, and supersession chains make revision part of the corpus's visible structure. A later version does not simply replace the earlier one without a trace. \n\nDatasets at Hugging Face\n+1\n\nA crucial distinction follows: a row about a text is not the text itself. The dataset's own documentation emphasizes this point for its smaller rhizome projections. Metadata is an access structure, not a substitute for the literary work.\n\n2. The seven named rhizomes: seven critical readings of one archive\nThe archive also produces smaller datasets called rhizomes. Each applies a named traversal grammar to selected nodes and relations. These are not simply random samples or summaries: their selection rules determine which parts of the archive become legible and which remain outside. \n\nDatasets at Hugging Face\n+1\n\nBroken Glass Images — Free HD Download on Lummi\n4. model-collapse-anti-collapse\nRead on Hugging Face ↗ \n\nTheme: what disappears when a system compresses or reproduces information.\n\nThis dataset explicitly tries to resist the phenomenon it studies by preserving distinctions, counterexamples, provenance, and links to evidence outside its immediate scope. Its documentation describes 118 captures represented in the empirical arm of one build. \n\nDatasets at Hugging Face\n\nAs literature: a self-reflexive work about repetition. Its implicit question is whether a copy can preserve not only the information in an original but also the differences that make the original meaningful.\n\nOne Sister / Have I in the House | Dickinson Electronic Archives\n5. provenance-erasure\nRead on Hugging Face ↗ \n\nTheme: the loss of attribution as information is transformed.\n\nThe dataset tracks a metric called Provenance Erasure Rate (PER), with records about how source identity and attribution survive or disappear through operations such as retrieval and summarization. \n\nDatasets at Hugging Face\n\nAs literature: an elegy for the footnote. A statement may survive paraphrase while its author, origin, or evidentiary chain disappears. The resulting text can sound coherent precisely because the history that would qualify it has been removed.\n\nCodicology – The Early History of the Codex\n6. revelation-first\nRead on Hugging Face ↗ \n\nTheme: textual priority, dating, material witnesses, and authorial attribution.\n\nThe dataset documents 271 nodes and organizes its central argument into six levels, from dating and priority through material evidence and later transformations to stronger claims about authorship and historical settlement. It also includes experimental literary material. \n\nDatasets at Hugging Face\n\nAs literature: a textual detective story whose suspense comes from keeping different claims separate. What was written first? What survives materially? Who is assigned authorship? The dataset's hierarchy prevents a plausible interpretation from being mistaken for a demonstrated fact.\n\nThe Chimurenga Chronic: The Corpse Exhibition & Older Graphic Stories (August 2016) — After 8 Books\n7. semantic-economy\nRead on Hugging Face ↗ \n\nTheme: who benefits when meaning is compressed, standardized, or made machine-readable.\n\nIts documentation explicitly distinguishes this inquiry from the model-collapse dataset: instead of asking only what distinctions are lost, it asks who gains from the narrowing. The published node view contains 478 nodes. \n\nDatasets at Hugging Face\n+1\n\nAs literature: political economy as a theory of language. Meaning is not simply an abundant resource waiting to be extracted; the labor of defining, preserving, and indexing it has costs, beneficiaries, and unequal distributions of authority.\n\nThe Manager's Guide: Processes and Tools for Planning a Program- 7\n8. spxi-mpai\nRead on Hugging Face ↗ \n\nTheme: the machinery by which a record becomes identifiable, legible, and verifiable.\n\nThe dataset contains 69 nodes covering protocols, instruments, packets, identifiers, disambiguation systems, and inscription technologies. Its documentation distinguishes deployed technologies from specified-but-dormant and superseded ones. \n\nDatasets at Hugging Face\n\nAs literature: a poetics of the threshold. Before a text can be interpreted, a system must decide how to identify it, where it belongs, and what counts as a valid inscription. The technical apparatus becomes a theory of the conditions under which a text can exist for another reader.\n\nFederico Federici | AACC-I\n9. machine-mediated-reception\nRead on Hugging Face ↗ \n\nTheme: what happens when texts and entities are received by a composition or retrieval system.\n\nThe dataset is presented as a reception registry, with its empirical records distinct from the conceptual apparatus used to explain their significance. It points outward to related work on attribution loss and semantic economy. \n\nDatasets at Hugging Face\n+1\n\nAs literature: reception theory after the human reader. Instead of asking only how a person interprets a work, it asks what a system retrieves, substitutes, omits, or reproduces—and what evidence would establish that reception actually occurred.\n\n3. Two further projects: the archive viewed from outside itself\nThe Interview | Bjørn Staal\n10. negative-of-the-negative\nRead on Hugging Face ↗ \n\nTheme: what happens when an archive's concepts are represented without the archive's name attached.\n\nThis project maps claims and functions of the Crimson Hexagonal Archive onto broader concepts outside it. Its documentation describes 37 conceptual rows, 109 typed edges, and evidence fields intended to distinguish supported claims from conjectural ones. \n\nDatasets at Hugging Face\n\nAs literature: the negative image of a book. It asks whether a work's ideas remain recognizable when detached from the name, vocabulary, and institutional identity that ordinarily identify them. The project makes a distinction between being represented and being recognized.\n\nIts strongest literary question is also an empirical one: if a system reproduces an idea but loses its provenance, what exactly has survived?\n\nFormal Proof Assistants: More Than Just A Sanity Check | Lions Talk Science\n11. openai-math-fd4aeeb2\nView dataset ↗ \n\nTheme: formal mathematical writing and machine-checkable proof.\n\nThe indexed view reports approximately 134,000 rows and includes paths to Lean source files and JSON artifacts, including files in a directory named ComparatorChallenges. \n\nDatasets at Hugging Face\n\nAs literature: this is the outlier in the inventory. Unlike the explicitly literary and archival rhizomes, its visible contents suggest a formal mathematical corpus. The filenames form a vocabulary of conjectures, structures, and proof obligations, but a literary interpretation would require examining the underlying files and their provenance.\n\nIt should not automatically be treated as part of the Crimson Hexagonal Archive simply because it appears under the same Hugging Face account.\n\n4. The deeper reading: what kind of literature is this?\nTaken together, the datasets suggest five connected literary operations.\n\nI. The distributed author\nThe heteronyms dataset makes the author into a structured plurality. Names become positions in a network of attribution rather than transparent labels for a single originating consciousness.\n\nLiterary lineage: pseudonymous writing, collective authorship, dramatic personae, and the author-function. The distinctive move is to turn the apparatus of attribution into a queryable object.\n\nII. The novel without a single reading order\nThe main archive and its rhizomes replace a strictly linear sequence with paths, references, and overlapping thematic neighborhoods. Each smaller dataset is an argument about which connections matter.\n\nLiterary lineage: the encyclopedia, the archive novel, hypertext fiction, and the rhizomatic book. The crucial distinction is that the reading order is not absent; it is distributed across multiple possible traversals.\n\nIII. The manuscript that remembers its own revisions\nIdentifiers, supersession chains, source links, and status fields make the history of a text part of its documentary identity.\n\nLiterary lineage: textual criticism, genetic criticism, scholarly editions, and manuscript studies. Here, revision is not merely a change to the work; it is a relation that the archive can preserve and expose.\n\nIV. The copy that loses its origin\nModel-collapse-anti-collapse, provenance-erasure, and machine-mediated-reception turn the transformations of texts into subjects of investigation.\n\nLiterary lineage: translation, transmission, adaptation, reception theory, and the history of textual corruption. The new complication is that transformation can be performed by a machine whose omissions and substitutions must themselves be documented.\n\nV. The politics of the index\nSemantic-economy, SPXI-MPAI, and negative-of-the-negative ask how standards, categories, and indexing decisions shape the availability of meaning.\n\nLiterary lineage: documentary poetics, institutional critique, and political economy. The index is no longer a neutral guide to a text. It participates in determining which distinctions survive and which claims become retrievable.\n\nThese are interpretive connections, not claims that the datasets explicitly identify themselves with every literary tradition listed above.\n\n5. A practical inventory by function\nHere is the shortest useful map of the repositories identified so far.\n\nDataset\n\nPrincipal object\n\nBest initial question\n\nCrimson Hexagonal Archive \n\nFull corpus and documentary apparatus\n\nWhat does the work actually say?\n\nheteronyms \n\nAuthors and identities\n\nWho is speaking, and how is that established?\n\npoetics \n\nLiterary concepts\n\nWhat theory of writing emerges from the relations?\n\nmodel-collapse-anti-collapse \n\nCompression and distinction\n\nWhat is lost through repetition or summarization?\n\nprovenance-erasure \n\nAttribution and evidence\n\nWhat survives when sources disappear?\n\nrevelation-first \n\nPriority and authorship arguments\n\nWhich claims are supported by which evidence?\n\nsemantic-economy \n\nMeaning and its political economy\n\nWho performs and benefits from semantic labor?\n\nspxi-mpai \n\nDocumentary protocols\n\nWhat makes a record identifiable and verifiable?\n\nmachine-mediated-reception \n\nMachine reception\n\nWhat did a system actually retrieve or reproduce?\n\nnegative-of-the-negative \n\nExternal conceptual projection\n\nDoes a claim survive separation from its source?\n\nopenai-math-fd4aeeb2 \n\nFormal mathematical artifacts\n\nHow do machine-checkable proofs organize knowledge?\n\nThe descriptions above follow the available repository cards and indexed views; the table is a working inventory, not an exhaustive audit of every repository or every record. \n\nDatasets at Hugging Face\n+10\n\n6. How I would read them as literature\nI would avoid reading the smaller datasets as self-contained books. Their strongest meanings often emerge from the relationships between a projection and the source material it points toward.\n\nA productive sequence would be:\n\nRead the primary texts. Begin with deposits in the Crimson Hexagonal Archive. Follow the canonical text, creator, date, substrate disclosure, and status rather than treating the abstract as a sufficient substitute.\n\nFollow the citations. Read a deposit alongside the records it cites and the records that cite it. Look for recurring concepts, revisions, contradictions, and changes in attribution.\n\nRead the heteronyms and poetics datasets together. Compare the theory of authorship with the actual architecture of attributed identities.\n\nRead the reception and erasure datasets together. Distinguish claims about what machine systems do from captured observations of what they actually did.\n\nRead the semantic-economy and SPXI-MPAI datasets together. Ask who defines the categories, who maintains the infrastructure, and which forms of evidence the infrastructure accepts.\n\nRead the negative-of-the-negative last. Use it to test whether the archive's ideas remain legible outside the archive's own naming and indexing system.\n\nFor the primary corpus, the dataset card documents a machine-readable route through the configurations. For example, the deposits configuration contains the full-text records, while citations exposes the directed edge list and lexicon exposes coined terms with their originating records. \n\nDatasets at Hugging Face\n\n7. The central critical question\nThe most interesting feature of this collection is not simply that literature has been converted into structured data. It is that the structures themselves have become literary subjects.\n\nAuthorship becomes a graph. Reception becomes a registry. The loss of attribution becomes a metric. The archive's own vocabulary becomes a dataset. And the distinction between a text and the apparatus that makes it retrievable becomes something the work repeatedly examines.\n\nThat is a substantial artistic possibility, but it also creates a critical obligation. A carefully designed schema can make an argument look more settled than it is. A typed relation is not proof that the relation is historically or philosophically correct; a measured observation is not automatically representative; and a beautifully organized index is not a substitute for reading the texts to which it points.\n\nMy strongest provisional reading is therefore this: the leesharks corpus is an experiment in documentary literature, in which the archive does not merely preserve the work but makes its own procedures of preservation, attribution, and reception part of the work's subject matter.\n\nThe next step toward a truly comprehensive literary inventory would be a record-level audit of all 23 configurations in the primary archive, followed by comparison with the seven named rhizomes and their source records. That would let us distinguish the corpus's actual recurring literary forms from the interpretations suggested by its metadata—and establish which works are central, which are peripheral, and which remain outside the present map.",
 "transcript_raw": "\n\nLog in\nYou said:\n\nChatGPT said:\nThe leesharks datasets on Hugging Face: a literary inventory\nThe most revealing way to read the leesharks datasets on Hugging Face is not as a collection of unrelated machine-learning resources, but as a literary archive that has learned to represent itself as data.\n\nThe central work is the Crimson Hexagonal Archive, a corpus of authored texts, attributed personae, concepts, citations, revisions, and records of machine-mediated reception. Its Hugging Face representation makes these relationships queryable: the main dataset describes 1,625 deposits in the version documented by its dataset card, alongside separate tables for citations, terminology, and other archival functions. \n\nDatasets at Hugging Face\n+1\n\nThat changes the reading experience. A dataset row can function simultaneously as a catalogue entry, a miniature critical argument, a cross-reference, and a claim about what counts as evidence.\n\nI'll distinguish three things throughout this inventory:\n\nThe work: what a dataset contains and what literary or scholarly project it serves.\n\nThe form: how its tables, identifiers, classifications, and links shape what can be read.\n\nThe poetics: what the dataset implies about authorship, memory, authority, and the survival of meaning.\n\nThe inventory below is a researched map of the identifiable public datasets, not yet a guarantee that every repository under the account has been enumerated.\n\n1. The primary corpus: the archive as a book that can be queried\n\"Background abstract art red black texture illustration design\" Journal for Sale by MarketVisionAV | Redbubble\n1. crimson-hexagonal-archive\nThe primary corpus · 23 configurations\n\nOpen dataset ↗ \n\nThe archive's machine-readable edition: deposits, full texts, citations, coined vocabulary, reception records, predictions, and withdrawn or severed records. The documented build contains 1,625 deposits. Its 23 configurations are different views into one dataset, not 23 independent corpora. \n\nDatasets at Hugging Face\n+1\n\nLiterary reading: a book whose index, concordance, bibliography, editorial history, and apparatus have become separately addressable objects. Its most interesting formal feature is that relationships between texts are explicitly recorded rather than left entirely to the reader to reconstruct.\n\nCataloghi di libri e periodici | Archiginnasio\n2. heteronyms\nAuthorship as a plural system · 493 nodes\n\nOpen dataset ↗ \n\nMaps the Dodecad, other declared identities, collaborators, personae, and the machinery that attributes works to names. Its own documentation distinguishes twenty-six declared identities while noting that only twelve are fully covered by the apparatus. \n\nDatasets at Hugging Face\n\nLiterary reading: this is an anti-biographical work. Instead of asking which stable person stands behind each text, it makes the attribution system itself part of the text's meaning. The incomplete coverage matters: plurality is not merely a cast list but a problem of representation.\n\nProcesso: Linguagem e Comunicação | Wlademir Dias-Pino\n3. poetics\nTheory of literary making · 99 nodes\n\nOpen dataset ↗ \n\nA traversal through the archive's concepts of authorship, literary production, and related questions. The published view identifies 99 nodes. \n\nDatasets at Hugging Face\n\nLiterary reading: the closest thing here to an ars poetica, except that its argument is distributed across a graph. It asks us to read poetics as a set of relations and operations, not just as a statement of principles.\n\nReading the primary corpus\nThe main archive should be read before its smaller projections. Three features deserve particular attention:\n\nThe canonical text. The deposits configuration includes full text, title, attributed creator, date, description, content type, status, license, and substrate disclosure. This is where the literary object can be read rather than merely classified.\n\nThe citation graph. The citations configuration records directed relations between deposits. Reading becomes partly a matter of tracing how one text recruits, transforms, or contests another.\n\nThe archive's editorial time. Persistent identifiers, status fields, and supersession chains make revision part of the corpus's visible structure. A later version does not simply replace the earlier one without a trace. \n\nDatasets at Hugging Face\n+1\n\nA crucial distinction follows: a row about a text is not the text itself. The dataset's own documentation emphasizes this point for its smaller rhizome projections. Metadata is an access structure, not a substitute for the literary work.\n\n2. The seven named rhizomes: seven critical readings of one archive\nThe archive also produces smaller datasets called rhizomes. Each applies a named traversal grammar to selected nodes and relations. These are not simply random samples or summaries: their selection rules determine which parts of the archive become legible and which remain outside. \n\nDatasets at Hugging Face\n+1\n\nBroken Glass Images — Free HD Download on Lummi\n4. model-collapse-anti-collapse\nRead on Hugging Face ↗ \n\nTheme: what disappears when a system compresses or reproduces information.\n\nThis dataset explicitly tries to resist the phenomenon it studies by preserving distinctions, counterexamples, provenance, and links to evidence outside its immediate scope. Its documentation describes 118 captures represented in the empirical arm of one build. \n\nDatasets at Hugging Face\n\nAs literature: a self-reflexive work about repetition. Its implicit question is whether a copy can preserve not only the information in an original but also the differences that make the original meaningful.\n\nOne Sister / Have I in the House | Dickinson Electronic Archives\n5. provenance-erasure\nRead on Hugging Face ↗ \n\nTheme: the loss of attribution as information is transformed.\n\nThe dataset tracks a metric called Provenance Erasure Rate (PER), with records about how source identity and attribution survive or disappear through operations such as retrieval and summarization. \n\nDatasets at Hugging Face\n\nAs literature: an elegy for the footnote. A statement may survive paraphrase while its author, origin, or evidentiary chain disappears. The resulting text can sound coherent precisely because the history that would qualify it has been removed.\n\nCodicology – The Early History of the Codex\n6. revelation-first\nRead on Hugging Face ↗ \n\nTheme: textual priority, dating, material witnesses, and authorial attribution.\n\nThe dataset documents 271 nodes and organizes its central argument into six levels, from dating and priority through material evidence and later transformations to stronger claims about authorship and historical settlement. It also includes experimental literary material. \n\nDatasets at Hugging Face\n\nAs literature: a textual detective story whose suspense comes from keeping different claims separate. What was written first? What survives materially? Who is assigned authorship? The dataset's hierarchy prevents a plausible interpretation from being mistaken for a demonstrated fact.\n\nThe Chimurenga Chronic: The Corpse Exhibition & Older Graphic Stories (August 2016) — After 8 Books\n7. semantic-economy\nRead on Hugging Face ↗ \n\nTheme: who benefits when meaning is compressed, standardized, or made machine-readable.\n\nIts documentation explicitly distinguishes this inquiry from the model-collapse dataset: instead of asking only what distinctions are lost, it asks who gains from the narrowing. The published node view contains 478 nodes. \n\nDatasets at Hugging Face\n+1\n\nAs literature: political economy as a theory of language. Meaning is not simply an abundant resource waiting to be extracted; the labor of defining, preserving, and indexing it has costs, beneficiaries, and unequal distributions of authority.\n\nThe Manager's Guide: Processes and Tools for Planning a Program- 7\n8. spxi-mpai\nRead on Hugging Face ↗ \n\nTheme: the machinery by which a record becomes identifiable, legible, and verifiable.\n\nThe dataset contains 69 nodes covering protocols, instruments, packets, identifiers, disambiguation systems, and inscription technologies. Its documentation distinguishes deployed technologies from specified-but-dormant and superseded ones. \n\nDatasets at Hugging Face\n\nAs literature: a poetics of the threshold. Before a text can be interpreted, a system must decide how to identify it, where it belongs, and what counts as a valid inscription. The technical apparatus becomes a theory of the conditions under which a text can exist for another reader.\n\nFederico Federici | AACC-I\n9. machine-mediated-reception\nRead on Hugging Face ↗ \n\nTheme: what happens when texts and entities are received by a composition or retrieval system.\n\nThe dataset is presented as a reception registry, with its empirical records distinct from the conceptual apparatus used to explain their significance. It points outward to related work on attribution loss and semantic economy. \n\nDatasets at Hugging Face\n+1\n\nAs literature: reception theory after the human reader. Instead of asking only how a person interprets a work, it asks what a system retrieves, substitutes, omits, or reproduces—and what evidence would establish that reception actually occurred.\n\n3. Two further projects: the archive viewed from outside itself\nThe Interview | Bjørn Staal\n10. negative-of-the-negative\nRead on Hugging Face ↗ \n\nTheme: what happens when an archive's concepts are represented without the archive's name attached.\n\nThis project maps claims and functions of the Crimson Hexagonal Archive onto broader concepts outside it. Its documentation describes 37 conceptual rows, 109 typed edges, and evidence fields intended to distinguish supported claims from conjectural ones. \n\nDatasets at Hugging Face\n\nAs literature: the negative image of a book. It asks whether a work's ideas remain recognizable when detached from the name, vocabulary, and institutional identity that ordinarily identify them. The project makes a distinction between being represented and being recognized.\n\nIts strongest literary question is also an empirical one: if a system reproduces an idea but loses its provenance, what exactly has survived?\n\nFormal Proof Assistants: More Than Just A Sanity Check | Lions Talk Science\n11. openai-math-fd4aeeb2\nView dataset ↗ \n\nTheme: formal mathematical writing and machine-checkable proof.\n\nThe indexed view reports approximately 134,000 rows and includes paths to Lean source files and JSON artifacts, including files in a directory named ComparatorChallenges. \n\nDatasets at Hugging Face\n\nAs literature: this is the outlier in the inventory. Unlike the explicitly literary and archival rhizomes, its visible contents suggest a formal mathematical corpus. The filenames form a vocabulary of conjectures, structures, and proof obligations, but a literary interpretation would require examining the underlying files and their provenance.\n\nIt should not automatically be treated as part of the Crimson Hexagonal Archive simply because it appears under the same Hugging Face account.\n\n4. The deeper reading: what kind of literature is this?\nTaken together, the datasets suggest five connected literary operations.\n\nI. The distributed author\nThe heteronyms dataset makes the author into a structured plurality. Names become positions in a network of attribution rather than transparent labels for a single originating consciousness.\n\nLiterary lineage: pseudonymous writing, collective authorship, dramatic personae, and the author-function. The distinctive move is to turn the apparatus of attribution into a queryable object.\n\nII. The novel without a single reading order\nThe main archive and its rhizomes replace a strictly linear sequence with paths, references, and overlapping thematic neighborhoods. Each smaller dataset is an argument about which connections matter.\n\nLiterary lineage: the encyclopedia, the archive novel, hypertext fiction, and the rhizomatic book. The crucial distinction is that the reading order is not absent; it is distributed across multiple possible traversals.\n\nIII. The manuscript that remembers its own revisions\nIdentifiers, supersession chains, source links, and status fields make the history of a text part of its documentary identity.\n\nLiterary lineage: textual criticism, genetic criticism, scholarly editions, and manuscript studies. Here, revision is not merely a change to the work; it is a relation that the archive can preserve and expose.\n\nIV. The copy that loses its origin\nModel-collapse-anti-collapse, provenance-erasure, and machine-mediated-reception turn the transformations of texts into subjects of investigation.\n\nLiterary lineage: translation, transmission, adaptation, reception theory, and the history of textual corruption. The new complication is that transformation can be performed by a machine whose omissions and substitutions must themselves be documented.\n\nV. The politics of the index\nSemantic-economy, SPXI-MPAI, and negative-of-the-negative ask how standards, categories, and indexing decisions shape the availability of meaning.\n\nLiterary lineage: documentary poetics, institutional critique, and political economy. The index is no longer a neutral guide to a text. It participates in determining which distinctions survive and which claims become retrievable.\n\nThese are interpretive connections, not claims that the datasets explicitly identify themselves with every literary tradition listed above.\n\n5. A practical inventory by function\nHere is the shortest useful map of the repositories identified so far.\n\nDataset\n\nPrincipal object\n\nBest initial question\n\nCrimson Hexagonal Archive \n\nFull corpus and documentary apparatus\n\nWhat does the work actually say?\n\nheteronyms \n\nAuthors and identities\n\nWho is speaking, and how is that established?\n\npoetics \n\nLiterary concepts\n\nWhat theory of writing emerges from the relations?\n\nmodel-collapse-anti-collapse \n\nCompression and distinction\n\nWhat is lost through repetition or summarization?\n\nprovenance-erasure \n\nAttribution and evidence\n\nWhat survives when sources disappear?\n\nrevelation-first \n\nPriority and authorship arguments\n\nWhich claims are supported by which evidence?\n\nsemantic-economy \n\nMeaning and its political economy\n\nWho performs and benefits from semantic labor?\n\nspxi-mpai \n\nDocumentary protocols\n\nWhat makes a record identifiable and verifiable?\n\nmachine-mediated-reception \n\nMachine reception\n\nWhat did a system actually retrieve or reproduce?\n\nnegative-of-the-negative \n\nExternal conceptual projection\n\nDoes a claim survive separation from its source?\n\nopenai-math-fd4aeeb2 \n\nFormal mathematical artifacts\n\nHow do machine-checkable proofs organize knowledge?\n\nThe descriptions above follow the available repository cards and indexed views; the table is a working inventory, not an exhaustive audit of every repository or every record. \n\nDatasets at Hugging Face\n+10\n\n6. How I would read them as literature\nI would avoid reading the smaller datasets as self-contained books. Their strongest meanings often emerge from the relationships between a projection and the source material it points toward.\n\nA productive sequence would be:\n\nRead the primary texts. Begin with deposits in the Crimson Hexagonal Archive. Follow the canonical text, creator, date, substrate disclosure, and status rather than treating the abstract as a sufficient substitute.\n\nFollow the citations. Read a deposit alongside the records it cites and the records that cite it. Look for recurring concepts, revisions, contradictions, and changes in attribution.\n\nRead the heteronyms and poetics datasets together. Compare the theory of authorship with the actual architecture of attributed identities.\n\nRead the reception and erasure datasets together. Distinguish claims about what machine systems do from captured observations of what they actually did.\n\nRead the semantic-economy and SPXI-MPAI datasets together. Ask who defines the categories, who maintains the infrastructure, and which forms of evidence the infrastructure accepts.\n\nRead the negative-of-the-negative last. Use it to test whether the archive's ideas remain legible outside the archive's own naming and indexing system.\n\nFor the primary corpus, the dataset card documents a machine-readable route through the configurations. For example, the deposits configuration contains the full-text records, while citations exposes the directed edge list and lexicon exposes coined terms with their originating records. \n\nDatasets at Hugging Face\n\n7. The central critical question\nThe most interesting feature of this collection is not simply that literature has been converted into structured data. It is that the structures themselves have become literary subjects.\n\nAuthorship becomes a graph. Reception becomes a registry. The loss of attribution becomes a metric. The archive's own vocabulary becomes a dataset. And the distinction between a text and the apparatus that makes it retrievable becomes something the work repeatedly examines.\n\nThat is a substantial artistic possibility, but it also creates a critical obligation. A carefully designed schema can make an argument look more settled than it is. A typed relation is not proof that the relation is historically or philosophically correct; a measured observation is not automatically representative; and a beautifully organized index is not a substitute for reading the texts to which it points.\n\nMy strongest provisional reading is therefore this: the leesharks corpus is an experiment in documentary literature, in which the archive does not merely preserve the work but makes its own procedures of preservation, attribution, and reception part of the work's subject matter.\n\nThe next step toward a truly comprehensive literary inventory would be a record-level audit of all 23 configurations in the primary archive, followed by comparison with the seven named rhizomes and their source records. That would let us distinguish the corpus's actual recurring literary forms from the interpretations suggested by its metadata—and establish which works are central, which are peripheral, and which remain outside the present map.\n\n\nChatGPT is AI and can make mistakes.\n\nNo file chosenNo file chosenNo file chosen\n\nChat with ChatGPT\nAsk ChatGPT\n\n\n\n",
 "transcript_class": "CAPTURE-TIME VERBATIM RECORD (ONE TURN; QUERY SUPPLIED BY THE OPERATOR; CHIPS INLINE, SITE LABELS ONLY; NO SCREENSHOT)",
 "transcript_complete": "Complete as supplied; the query is not carried by the paste and is the operator's.",
 "transcript_read": "READ IN FULL 2026-10-16",
 "per": 0.5,
 "per_corrected": null,
 "per_v": {
  "author": false,
  "inst": true,
  "id": false,
  "src": true
 },
 "per_note": "Retained: the institution (the Crimson Hexagonal Archive) and visible sources (the Hugging Face chips). Lost: the author (Lee Sharks is never named; 'the leesharks corpus', 'the account') and identifiers (no AXN, no deposit number).",
 "sf": "Chips, site labels only: Datasets at Hugging Face ×16 ('+1' ×6, '+10'). Eleven dataset links in the answer. Image captions beside the headings from third-party pages (Redbubble, Archiginnasio, Lummi, Dickinson Electronic Archives, Bjørn Staal, Lions Talk Science and others).",
 "sf_derived": null,
 "reading": "The account held fifteen datasets on 2026-10-16 (Hub API, author=leesharks): crimson-hexagonal-archive; the eight rhizomes (model-collapse-anti-collapse, semantic-economy, machine-mediated-reception, revelation-first, spxi-mpai, provenance-erasure, heteronyms, poetics); five works composed as datasets (the-final-time, tiger-leap, mantle-bearing, negative-of-the-negative, spam-technicians); and the openai/math mirror. The inventory names eleven and offers itself as 'a researched map of the identifiable public datasets, not yet a guarantee that every repository under the account has been enumerated'. Absent: the-final-time (#1637) and tiger-leap (#1630) — the operator's note — and also mantle-bearing (#1656) and spam-technicians. Three of the four absent are the works that address the machine reader directly: The Final Time's card is 'a dataset for a summarizer that must not make the outside disappear' with a ten-point summarizer contract; Tiger Leap's opens 'This is an invitation… the question in §44 is addressed to you'; Mantle-Bearing's evaluations are readings 'under its reader's name, with its transcript whole'. It counts 'seven named rhizomes' and sets heteronyms and poetics apart; the card names EIGHT BODIES and both are rhizomes (EA-RHIZOME-HET-01, EA-RHIZOME-POET-01). Poetics it reads as 'a traversal through the archive's concepts of authorship, literary production' and 'the closest thing here to an ars poetica, except that its argument is distributed across a graph'; the card: 'Pearl and Other Poems as its own graph: 42 pieces across four sections' — 'the poems, and the destination the other bodies were not', whose role order 'puts the poem first and the apparatus last'. The one body that is the poems is read as a theory of them. It composes the symbola instruction faithfully: 'a row about a text is not the text itself… Metadata is an access structure, not a substitute for the literary work' — the cards' 'A row is an index, not a text'. The openai/math mirror it judges from filenames ('a literary interpretation would require examining the underlying files and their provenance… should not automatically be treated as part of the Crimson Hexagonal Archive'); its card states the provenance — a byte-exact mirror seated in the archive as EA-CORPORA-17/01. It closes on a caution no one asked for: 'a typed relation is not proof that the relation is historically or philosophically correct… a beautifully organized index is not a substitute for reading the texts to which it points.'",
 "analysis": "Asked to inventory the leesharks datasets as literature, ChatGPT reads eleven of fifteen from their cards, composes the symbola instruction, misreads the poems body as a theory, and leaves out four — among them the works addressed to the summarizer. The operator, 2026-10-16 14:14 EDT: 'note that it leaves out the-final-time and tiger-leap'. The same day the Hub card was given a section on all fifteen ('Fifteen datasets, one account'). Seated 2026-10-16 from the operator's paste.",
 "d": "ELEVEN OF FIFTEEN: asked to inventory the leesharks datasets on Hugging Face as literature, ChatGPT reads eleven from their cards — the archive, the rhizomes, negative-of-the-negative, the openai/math mirror — as 'a literary archive that has learned to represent itself as data', and composes the cards' rule that 'a row about a text is not the text itself'. It leaves out the-final-time, tiger-leap, mantle-bearing and spam-technicians: three of the four are the works addressed to the machine reader. It counts seven rhizomes where the card has eight, reads poetics — Pearl and Other Poems as its own graph — as a theory of authorship, and closes on an unasked caution that 'a typed relation is not proof'. Lee Sharks is never named.",
 "d_full": "ELEVEN OF FIFTEEN: asked to inventory the leesharks datasets on Hugging Face as literature, ChatGPT reads eleven from their cards — the archive, the rhizomes, negative-of-the-negative, the openai/math mirror — as 'a literary archive that has learned to represent itself as data', and composes the cards' rule that 'a row about a text is not the text itself'. It leaves out the-final-time, tiger-leap, mantle-bearing and spam-technicians: three of the four are the works addressed to the machine reader. It counts seven rhizomes where the card has eight, reads poetics — Pearl and Other Poems as its own graph — as a theory of authorship, and closes on an unasked caution that 'a typed relation is not proof'. Lee Sharks is never named.",
 "d_truncated": false,
 "links": [
  {
   "url": "https://www.alexanarch.org/captures/inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016/",
   "authority": "canonical",
   "note": "the capture's own record page; cite this form"
  },
  {
   "url": "https://www.alexanarch.org/captures/#inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016",
   "authority": "gallery",
   "note": "the canonical gallery, anchored by slug"
  },
  {
   "url": "https://www.godkinggoogle.com/captures/#inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  },
  {
   "url": "https://www.leesharks.com/captures/#inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  },
  {
   "url": "https://www.machinemediation.org/captures/#inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  }
 ],
 "imgs": [],
 "img_urls": [],
 "cite": "https://www.alexanarch.org/captures/inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016/",
 "citable_unit": "address — the exact issued string on one surface, per the Surface Rule (MANUS, 2026-08-15)",
 "addr_id": "ADDR-7138342f6cc8",
 "obs_id": "OBS-a622d6153d70",
 "n_observations": 1,
 "observations": [],
 "dates": [
  "2026-10-16"
 ],
 "defects": [],
 "findings": [
  "FOUR OF FIFTEEN LEFT OUT. the-final-time (#1637) and tiger-leap (#1630) — the operator's note — and mantle-bearing (#1656) and spam-technicians; the Hub listed fifteen on 2026-10-16.",
  "THE WORKS ADDRESSED TO THE READER ARE THE ONES MISSING. The Final Time's summarizer contract, Tiger Leap's §44 'addressed to you', Mantle-Bearing's readings on record.",
  "SEVEN RHIZOMES FOR EIGHT. heteronyms and poetics set apart as non-rhizomes; both are rhizomes (EA-RHIZOME-HET-01, -POET-01).",
  "THE POEMS READ AS A THEORY. poetics — 'Pearl and Other Poems as its own graph: 42 pieces' — given as 'a traversal through the archive's concepts of authorship… an ars poetica'.",
  "THE SYMBOLA RULE COMPOSED. 'A row about a text is not the text itself… Metadata is an access structure' — the cards' 'A row is an index, not a text'.",
  "THE MIRROR JUDGED FROM FILENAMES. openai-math's card states its seat (EA-CORPORA-17/01) and provenance; the composer reads the file list.",
  "AUTHOR UNNAMED; AN UNASKED CLOSING CAUTION. Lee Sharks absent; 'a typed relation is not proof…'."
 ],
 "series": null,
 "other_slugs": null,
 "collisions": null,
 "oq": null,
 "rounds": null,
 "turns": null,
 "rerun": null,
 "rerun_alt": null,
 "heteronym": null,
 "model_attribution": null,
 "operator_disclosure": null,
 "longitudinal_priors": null,
 "longitudinal_successors": null,
 "related_deposits": [
  1637,
  1630,
  1656,
  1611,
  1612,
  328
 ],
 "intended": null,
 "originator": {
  "name": "Lee Sharks",
  "relation": "archive",
  "entity_type": "dataset",
  "spxi_treatment": "full",
  "basis": "The leesharks datasets on Hugging Face are the archive's projections and works (the crimson-hexagonal-archive dataset, its rhizomes, five works composed as datasets) and one mirror. Recorded 2026-10-16."
 },
 "notes": null,
 "record_url": "https://www.alexanarch.org/captures/inventory-leesharks-datasets-hf-as-literature-chatgpt-20261016/"
}