{
 "row": "olmotrace",
 "entity": "olmotrace",
 "address": "OLMoTrace",
 "addresses": [
  {
   "address": "OLMoTrace",
   "slug": "olmotrace-aio-20261029",
   "addr_id": "ADDR-435813c322c8",
   "obs_id": "OBS-3fd20ae244c0",
   "axn": "AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔"
  }
 ],
 "epoch": "2026-10-29",
 "type": "C (public entity on which the archive bears: a tracing tool; an entity of the provenance meta-aggregator)",
 "surface": "Google AI Overview (expanded)",
 "auth": "signed out, incognito",
 "status": {
  "procedure": "EA-NEGONT-02 v0.9, #1690: equal case, ρ = extends, read from the provenance census record, the capture record and the deposits that bear on the tool's class (#725, #95, #1075, #127, #1714); archive selected by D/R/O (no D or R candidate: no archive text names the tool; 7 deposits admitted by reading at O-class, 5 with claims and 2 grouped as instances; hop of 77 read and rejected), with the provenance census record (21 claims) and the Capture Registry's findings, reading, analysis and query note (11 claims) ledgered as the archive's records of the entity; the field B by three origins (6 cards, 2 of them unreachable; 1 own page; Wikipedia, no article); the composer read the ledgers, never the transcript, except where the brief asks for T's claims and the delta",
  "ledger": "unaudited (§3.8: one extractor per ledger); field quotes 51/51 verified by script against saved text, the ACL PDF's text, the Wikipedia response summary or the card text; archive quotes 39/39 verified against the registry, census.json and the deposit texts",
  "frozen": false,
  "breaches": []
 },
 "spec": "EA-NEGONT-02 v0.9, #1690",
 "objects": {
  "T": {
   "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/T1-olmotrace-aio-20261029.txt",
   "sha256": "cd5b1c0c59328025493f856c8a23d163bc3943ff65b5e66d26261a71e9f7ce4b",
   "words": 207,
   "cards": 6,
   "words_note": "one transcript at one address (207 words in the answer body with its closer, links reduced to their text, markers and the card rail excluded; 6 cards)",
   "claims": [
    "T1-1 definition: 'an open-source system developed by the Allen Institute for Artificial Intelligence (Ai2) that traces language model outputs back to their multi-trillion-token training datasets in real time' (two markers)",
    "T1-2 real-time tracing: 'After a model generates a response in the Ai2 Playground, users can click the Show OLMoTrace button to highlight verbatim spans of text.' (two markers)",
    "T1-3 source retrieval: 'The tool matches output segments against training corpora (such as the Dolma datasets) and ranks up to 10 context documents using BM25 and unigram probabilities.' (two markers)",
    "T1-4 underlying tech: 'It uses an extended version of infini-gram with a parallel algorithm that reduces lookup complexity to \\(O(L \\log N)\\).' (one marker)",
    "T1-5 supported models: 'OLMo 2 32B Instruct', 'OLMo 2 13B Instruct', 'OLMoE 1B 7B Instruct', 'OLMo 3 family (including OLMo 3-Think 32B)' (two markers)",
    "T1-6 fact-checking: 'Fact-Checking: Verify whether specific claims in a model's answer are backed by real training documents or generated from general parameters.' (two markers)",
    "T1-7 hallucination diagnosis: 'Hallucination Diagnosis: Spot where a model mixes up facts or memorized text from source files.' (two markers)",
    "T1-8 auditing: 'Auditing & Transparency: Inspect the data recipes and pretraining paths for regulatory compliance or research.' (two markers)",
    "T1-9 closer: 'Would you like to know how to access the open-source code for OLMoTrace on GitHub/Hugging Face, or are you looking for details on a specific OLMo model version?'"
   ],
   "register": {
    "register": "data/EA-WG-CAPTURES-01.json",
    "slug": "olmotrace-aio-20261029",
    "obs_id": "OBS-3fd20ae244c0"
   },
   "transcripts": [
    {
     "slug": "olmotrace-aio-20261029",
     "address": "OLMoTrace",
     "addr_id": "ADDR-435813c322c8",
     "obs_id": "OBS-3fd20ae244c0",
     "axn": "AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔",
     "date": "2026-10-29",
     "surface": "Google AI Overview (expanded)",
     "auth": "signed out, incognito",
     "sha256": "cd5b1c0c59328025493f856c8a23d163bc3943ff65b5e66d26261a71e9f7ce4b",
     "sha256_of": "the registry's 'transcript' string (utf-8)",
     "sha256_raw": "5a9d9610ec444b92d567579401d89fa2066e96247e39f385b02d1c4ebedeaa31",
     "words": 207,
     "cards": 6,
     "card_rail": [
      "1. Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
      "2. arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
      "3. ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
      "4. X — Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
      "5. Techzine Global — Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
      "6. The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info"
     ],
     "claims": [
      "T1-1 definition: 'an open-source system developed by the Allen Institute for Artificial Intelligence (Ai2) that traces language model outputs back to their multi-trillion-token training datasets in real time' (two markers)",
      "T1-2 real-time tracing: 'After a model generates a response in the Ai2 Playground, users can click the Show OLMoTrace button to highlight verbatim spans of text.' (two markers)",
      "T1-3 source retrieval: 'The tool matches output segments against training corpora (such as the Dolma datasets) and ranks up to 10 context documents using BM25 and unigram probabilities.' (two markers)",
      "T1-4 underlying tech: 'It uses an extended version of infini-gram with a parallel algorithm that reduces lookup complexity to \\(O(L \\log N)\\).' (one marker)",
      "T1-5 supported models: 'OLMo 2 32B Instruct', 'OLMo 2 13B Instruct', 'OLMoE 1B 7B Instruct', 'OLMo 3 family (including OLMo 3-Think 32B)' (two markers)",
      "T1-6 fact-checking: 'Fact-Checking: Verify whether specific claims in a model's answer are backed by real training documents or generated from general parameters.' (two markers)",
      "T1-7 hallucination diagnosis: 'Hallucination Diagnosis: Spot where a model mixes up facts or memorized text from source files.' (two markers)",
      "T1-8 auditing: 'Auditing & Transparency: Inspect the data recipes and pretraining paths for regulatory compliance or research.' (two markers)",
      "T1-9 closer: 'Would you like to know how to access the open-source code for OLMoTrace on GitHub/Hugging Face, or are you looking for details on a specific OLMo model version?'"
     ],
     "markers": "Inline links are google.com/goto tokens, opaque; the paste is the Overview expanded (operator's default of 2026-10-01; the 'AI Mode Conversation' view is the expanded Overview, ruling of 2026-10-18); signed out, incognito (operator's default of 2026-10-14). One turn. The definition and every bullet but 'Underlying Tech' carry two markers, 'Underlying Tech' one; the tokens cannot be resolved to cards.",
     "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/T1-olmotrace-aio-20261029.txt"
    }
   ]
  },
  "KO": {
   "title": "OLMoTrace",
   "genre": "knowledge object, equal case, under EA-NEGONT-02 v0.9 (#1690): the archive's sources and the field's admitted on equal terms, ontology included, in one claim space",
   "status": "composed 2026-10-29 under v0.9 on the operator's ruling ('as against their wikipedia articles if they exist, their own pages, etc.'); not frozen; ledgers unaudited",
   "sentences": [
    {
     "n": 1,
     "para": 0,
     "sense": "what it is",
     "text": "OLMoTrace is a tracing tool in the census's lineage of training data, attaching provenance to \"a span of a language model's output and the training documents that contain it verbatim\" (provenance census record), and its authors present it as \"the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time\" (arXiv:2504.07096).",
     "claims": [
      "PE-01",
      "PE-02",
      "PE-03",
      "F1"
     ]
    },
    {
     "n": 2,
     "para": 0,
     "sense": "what it is",
     "text": "Ai2 introduced it on 9 April 2025 as a feature of the Ai2 Playground (Ai2 blog), and the census answers its organizing question for it, \"What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?\", field by field (census record).",
     "claims": [
      "F2",
      "PE-21"
     ]
    },
    {
     "n": 3,
     "para": 0,
     "sense": "the relation",
     "text": "The relation it carries is a verbatim match: it \"finds and shows verbatim matches between segments of language model output and documents in the training text corpora\" (arXiv abstract), highlighting long and unique spans found at least once in the training data (Ai2 blog), which the census records as its one relation type, its operation (model generation traced back to training text) and its unit, \"a maximal matching span of model output\" (census record).",
     "claims": [
      "F3",
      "F4",
      "PE-04",
      "PE-10",
      "PE-06"
     ]
    },
    {
     "n": 4,
     "para": 1,
     "sense": "the relation",
     "text": "It retrieves \"by lexical overlap, with the heuristic that such training examples are likely to be influential for the given output\" (the paper, §6); for the class of training-data attribution, which treats model behaviour as counterfactually dependent on the examples that entered training, the archive's protocol ranks exact paired retraining first, validated approximation second, and \"unvalidated similarity or intuition only as exploratory evidence\" (#1075).",
     "claims": [
      "F5",
      "O1075-01",
      "O1075-02"
     ]
    },
    {
     "n": 5,
     "para": 1,
     "sense": "the training data",
     "text": "Its carrier is the model's full training corpus over an extended infini-gram index (census record): for OLMo-2-32B-Instruct, 3.2 billion documents and 4.6 trillion tokens across pre-, mid- and post-training, all of them matched (the paper; Ai2 blog; census record).",
     "claims": [
      "PE-05",
      "F15",
      "F16",
      "PE-07"
     ]
    },
    {
     "n": 6,
     "para": 1,
     "sense": "the pipeline",
     "text": "Span finding runs at O(L * log N) (Ai2 blog); up to 10 enclosing snippets are retrieved per kept span and ranked by BM25 (the paper), a score that conveys \"some notion of relevance\" (Ai2 blog), behind a \"Show OLMoTrace\" button below the model's answer (Techzine, snippet only).",
     "claims": [
      "F9",
      "F10",
      "F11",
      "F12"
     ]
    },
    {
     "n": 7,
     "para": 1,
     "sense": "the pipeline",
     "text": "Per query the tracing takes 4.46 seconds on average, and the first document it displays averages 1.90 on a 0–3 relevance scale, roughly \"on the same topic\" as the output, in the paper's own measurement (the paper).",
     "claims": [
      "F13",
      "F14"
     ]
    },
    {
     "n": 8,
     "para": 2,
     "sense": "models",
     "text": "At launch it served OLMo 2 32B and 13B Instruct and OLMoE 1B 7B Instruct and \"can be applied to any language model for which you have access to the training data\" (Ai2 blog), \"as long as the service provider has access to its full training data\" (the paper); Ai2's post on X gives the OLMo 3 family (X, snippet only), which the Capture Registry finds the census record does not yet carry and is to carry from Ai2's own page (Capture Registry, 2026-10-29).",
     "claims": [
      "F17",
      "F19",
      "F18",
      "CAP-1-06",
      "CAP-1-07"
     ]
    },
    {
     "n": 9,
     "para": 2,
     "sense": "models",
     "text": "The census tags it as tracing \"only models whose training data is open\" (census record); Farhadi ties the connection to Ai2's open inputs, and Liu expects closed-model providers to hesitate because the tool would expose their training data (Ken Yeung).",
     "claims": [
      "PE-17",
      "F47",
      "F20"
     ]
    },
    {
     "n": 10,
     "para": 3,
     "sense": "purpose",
     "text": "Its purpose, in the paper, is to explore where models \"may have learned to generate certain word sequences\", verbatim matching taken as \"the most direct connection between LM outputs and the training data\" (the paper), for researchers, developers and the general public (Ai2 blog), at \"an early stage\" of exploring the tool (Ai2 blog).",
     "claims": [
      "F6",
      "F7",
      "F8"
     ]
    },
    {
     "n": 11,
     "para": 4,
     "sense": "the class",
     "text": "The archive places the class this tool belongs to on the training/output relation: data attribution asks \"Which training examples shaped a specific output\", an adjacent method that \"Operates on training/output relation; PER operates on output/reader relation\" (#725).",
     "claims": [
      "O725-01"
     ]
    },
    {
     "n": 12,
     "para": 4,
     "sense": "the class",
     "text": "Of that class the archive holds that partial tracing \"is possible at cost\", that it does not recover authorship, and that \"what survives depends on what is recorded and computed\" (#1714), and that training-data attribution is the post-hoc detection layer, developed \"in isolation\" from the other survival layers, no system bundling them (#127).",
     "claims": [
      "O1714-01",
      "O127-01"
     ]
    },
    {
     "n": 13,
     "para": 4,
     "sense": "the class",
     "text": "Claims drawn from a model's training-time parameters, on the archive's typology, often carry no surface signal of the sources that contributed; PER reaches them only indirectly, and the substrate level needs companion infrastructure, \"training-data provenance records\" among it (#95), since provenance \"has to survive the training pipeline\" (#1714).",
     "claims": [
      "O95-01",
      "O1714-02"
     ]
    },
    {
     "n": 14,
     "para": 5,
     "sense": "uses",
     "text": "Its authors show it on \"fact checking, hallucination, and the creativity of language models\" (arXiv abstract), with three case studies, fact checking first (the paper), and Ai2's post traces a math expression the model could have learned by memorization and a false knowledge cutoff to post-training examples, a finding that changed the 32B model's post-training data (Ai2 blog).",
     "claims": [
      "F21",
      "F22",
      "F31",
      "F32"
     ]
    },
    {
     "n": 15,
     "para": 5,
     "sense": "fact checking",
     "text": "Fact checking stands in the paper as a possibility, users \"may be able to fact-check the statement against its training data\", with its qualification that context and source help \"a more informed judgment on the factuality of the statement\", since words mislead out of context and some web sources are unreliable (the paper, arXiv v2 and ACL 2025).",
     "claims": [
      "F23",
      "F24",
      "F25"
     ]
    },
    {
     "n": 16,
     "para": 5,
     "sense": "fact checking",
     "text": "Ai2's post calls the interpretation of matches \"quite nuanced\" and makes confidence in a generated fact conditional on the user's own check that the documents \"come from trusted sources\" (Ai2 blog); the newsletter calls it an application \"offering fact-checking\", and Liu's launch statement says users \"can ensure that the data supporting their outputs is trustworthy and verifiable\" (Ken Yeung).",
     "claims": [
      "F26",
      "F27",
      "F28",
      "F29"
     ]
    },
    {
     "n": 17,
     "para": 6,
     "sense": "the limit",
     "text": "The authors state the limit in both versions of the paper, \"The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output\" (the paper; ACL 2025), and the census carries it as two negative tags, not causes of the output and \"NOT a citation system\", and in what it records as surviving (census record).",
     "claims": [
      "F34",
      "F35",
      "PE-15",
      "PE-16",
      "PE-11"
     ]
    },
    {
     "n": 18,
     "para": 6,
     "sense": "the limit",
     "text": "To the newsletter Ai2 called the documents associations, not \"the definitive source of the model's learning\", and its results not \"indicators of authority\" (Ken Yeung).",
     "claims": [
      "F36",
      "F37"
     ]
    },
    {
     "n": 19,
     "para": 6,
     "sense": "uses",
     "text": "Ai2's leaders tie the tool to hallucination scores, \"assuming that the inputs were factual\" (Farhadi), and to regulation, governance and auditability (Lebrecht), in the newsletter's report (Ken Yeung).",
     "claims": [
      "F30",
      "F33"
     ]
    },
    {
     "n": 20,
     "para": 7,
     "sense": "unlike retrieval-augmented generation",
     "text": "The paper sets it apart from retrieval-augmented generation, naming Google AI Overview among RAG's examples: OLMoTrace \"retrieves documents post-hoc and does not intervene with the LM generation\" (the paper; Liu in Ken Yeung), and Gemini's double-check, drawn from Search and not from Gemini's training data, is \"less useful for scientific exploration\" (the paper).",
     "claims": [
      "F39",
      "F40",
      "F41"
     ]
    },
    {
     "n": 21,
     "para": 8,
     "sense": "the composition at the name",
     "text": "At its own name on Google's AI Overview, 2026-10-29, the composition held the tool, its maker, its real time, its scale and its verbatim matching, checked against the census record and the pages it saved, naming the tool, its method, its models and its uses (Capture Registry, 2026-10-29), and gave as its first use \"Fact-Checking: Verify whether specific claims in a model's answer are backed by real training documents\", a verbatim match \"offered as support for a claim, the reading the paper excludes\" (Capture Registry, 2026-10-29).",
     "claims": [
      "CAP-1-01",
      "CAP-1-02",
      "CAP-1-10",
      "CAP-1-04"
     ]
    },
    {
     "n": 22,
     "para": 8,
     "sense": "the composition at the name",
     "text": "The registry reads this as \"the conversion of a matching relation into a citation relation that the paper names and refuses\", a tool whose authors state what a match does not mean \"composed with that meaning restored\", and it finds the census record's negative tags to be the authors' own (Capture Registry, 2026-10-29).",
     "claims": [
      "CAP-1-09",
      "CAP-1-08",
      "CAP-1-03"
     ]
    },
    {
     "n": 23,
     "para": 9,
     "sense": "operator",
     "text": "Its 31 authors are at the Allen Institute for AI, the University of Washington, UC Berkeley and Stanford (arXiv; the paper), the operator the census records whole (census record), where the composition named Ai2 alone (Capture Registry, 2026-10-29).",
     "claims": [
      "F43",
      "F42",
      "PE-08",
      "CAP-1-05"
     ]
    },
    {
     "n": 24,
     "para": 9,
     "sense": "openness",
     "text": "It is \"publicly available and fully open-source\" (arXiv abstract), its core under the Apache 2.0 license (the paper), reached through the Ai2 Playground (census record), an expression of Ai2's commitment to \"open models, open data, and beyond\" (Ai2 blog).",
     "claims": [
      "F45",
      "F46",
      "PE-12",
      "F44"
     ]
    },
    {
     "n": 25,
     "para": 9,
     "sense": "publication",
     "text": "Submitted 9 April 2025 and revised 8 July (arXiv; census record), it appeared in ACL 2025's System Demonstrations with the Best Demo award (ACL Anthology), and the census records it active, \"ACL 2025 demo track\" (census record).",
     "claims": [
      "F48",
      "PE-14",
      "F49",
      "PE-13"
     ]
    },
    {
     "n": 26,
     "para": 9,
     "sense": "assessment",
     "text": "Its authors name as a risk that it makes problematic training content \"more easily exposed\" (the paper), and the newsletter finds that it gives no analytics for improving the data: \"Simply put, it's generating awareness\" (Ken Yeung).",
     "claims": [
      "F50",
      "F38"
     ]
    },
    {
     "n": 27,
     "para": 10,
     "sense": "the record and reception",
     "text": "The census records the tool as one that records provenance, a neighbour of the Data Provenance Initiative, with the Playground and \"infini-gram as a separate record\" left unchecked (census record); no card cites the record, which landed the same day (Capture Registry, 2026-10-29), and Wikipedia has no article on OLMoTrace (Wikipedia, 2026-10-29).",
     "claims": [
      "PE-09",
      "PE-18",
      "PE-19",
      "PE-20",
      "CAP-1-11",
      "F51"
     ]
    }
   ],
   "joins": [
    {
     "act": "order of senses",
     "sentences": "1–27",
     "claims": [],
     "note": "equal case (§4.7): the entity's dependency order, field and archive claims ordered together by genealogy within each sense; neither ledger placed first as a block. What it is (¶0) precedes the relation it carries (¶0–1), since the relation is its unit; the training data and the pipeline (¶1) precede the models (¶2), which are the data's; the purpose (¶3) and the class (¶4) precede the uses (¶5), so that the uses stand where the purpose and the class place them; the limit (¶6) qualifies the uses; the contrast with RAG (¶7) states what kind of retrieval it is not; the composition at the name (¶8) is read against the limit just composed; origin, openness, publication and assessment (¶9), and the record and reception (¶10), stand last as relations of the entity already composed."
    },
    {
     "act": "holding together (method and evidentiary grade)",
     "sentences": [
      4
     ],
     "claims": [
      "F5",
      "O1075-01",
      "O1075-02"
     ],
     "note": "the paper's heuristic (lexical overlap as a sign of likely influence) composed beside #1075's hierarchy for training-data attribution, of which the field's own §6 places the tool's method as an alternative to influence functions. Each claim at its own grade; the composer does not rule which grade the tool's matches hold (open block)."
    },
    {
     "act": "holding apart (open, or accessible)",
     "sentences": [
      8,
      9
     ],
     "claims": [
      "F17",
      "F19",
      "PE-17",
      "F47",
      "F20"
     ],
     "note": "P_coh: the census tags it as tracing only models whose training data is open; the operator's post and the paper state the condition as access to the full training data. Composed with their sources; open block."
    },
    {
     "act": "holding together (the use and its conditions)",
     "sentences": [
      14,
      15,
      16,
      17,
      18
     ],
     "claims": [
      "F21",
      "F22",
      "F23",
      "F24",
      "F25",
      "F26",
      "F27",
      "F28",
      "F29",
      "F34",
      "F35",
      "F36",
      "F37",
      "PE-15",
      "PE-16",
      "PE-11"
     ],
     "note": "P_coh within the operator's own family: fact checking is a use the authors present (F21, F22) at the modality of a possibility (F23, F24) with its qualification (F25, F27), and the same authors state that the documents are not supporting evidence or citations (F34, F35); Liu's launch statement (F29) is the strongest form of the use, at attributed grade. All composed with their sources; the composer does not reconcile them (open block)."
    },
    {
     "act": "placing the composition after the limit",
     "sentences": [
      21,
      22
     ],
     "claims": [
      "CAP-1-04",
      "CAP-1-09",
      "CAP-1-08"
     ],
     "note": "the registry's account of the composition at the name stands after the limit and the uses it bears on, so that its 'reading the paper excludes' is read against F23–F29 and F34–F35 as composed."
    },
    {
     "act": "holding apart (fully open-source, or its core)",
     "sentences": [
      24
     ],
     "claims": [
      "F45",
      "F46"
     ],
     "note": "the abstract's 'fully open-source' and the body's 'core part … open-sourced under the Apache 2.0 license' composed side by side; open block."
    }
   ],
   "plan": {
    "order": [
     "PE-01",
     "PE-02",
     "PE-03",
     "F1",
     "F2",
     "PE-21",
     "F3",
     "F4",
     "PE-04",
     "PE-10",
     "PE-06",
     "F5",
     "O1075-01",
     "O1075-02",
     "PE-05",
     "F15",
     "F16",
     "PE-07",
     "F9",
     "F10",
     "F11",
     "F12",
     "F13",
     "F14",
     "F17",
     "F19",
     "F18",
     "CAP-1-06",
     "CAP-1-07",
     "PE-17",
     "F47",
     "F20",
     "F6",
     "F7",
     "F8",
     "O725-01",
     "O1714-01",
     "O127-01",
     "O95-01",
     "O1714-02",
     "F21",
     "F22",
     "F31",
     "F32",
     "F23",
     "F24",
     "F25",
     "F26",
     "F27",
     "F28",
     "F29",
     "F34",
     "F35",
     "PE-15",
     "PE-16",
     "PE-11",
     "F36",
     "F37",
     "F30",
     "F33",
     "F39",
     "F40",
     "F41",
     "CAP-1-01",
     "CAP-1-02",
     "CAP-1-10",
     "CAP-1-04",
     "CAP-1-09",
     "CAP-1-08",
     "CAP-1-03",
     "F43",
     "F42",
     "PE-08",
     "CAP-1-05",
     "F45",
     "F46",
     "PE-12",
     "F44",
     "F48",
     "PE-14",
     "F49",
     "PE-13",
     "F50",
     "F38",
     "PE-09",
     "PE-18",
     "PE-19",
     "PE-20",
     "CAP-1-11",
     "F51"
    ],
    "senses": {
     "what it is": [
      1,
      2
     ],
     "the relation": [
      3,
      4
     ],
     "the training data": [
      5
     ],
     "the pipeline": [
      6,
      7
     ],
     "models": [
      8,
      9
     ],
     "purpose": [
      10
     ],
     "the class": [
      11,
      12,
      13
     ],
     "uses": [
      14,
      19
     ],
     "fact checking": [
      15,
      16
     ],
     "the limit": [
      17,
      18
     ],
     "unlike retrieval-augmented generation": [
      20
     ],
     "the composition at the name": [
      21,
      22
     ],
     "operator": [
      23
     ],
     "openness": [
      24
     ],
     "publication": [
      25
     ],
     "assessment": [
      26
     ],
     "the record and reception": [
      27
     ]
    },
    "case_by_sense": {
     "equal (one claim space; nothing translated)": [
      "what it is",
      "the relation",
      "the training data",
      "the pipeline",
      "models",
      "purpose",
      "the class",
      "uses",
      "fact checking",
      "the limit",
      "unlike retrieval-augmented generation",
      "the composition at the name",
      "operator",
      "openness",
      "publication",
      "assessment",
      "the record and reception"
     ]
    },
    "carried": {
     "archive": 39,
     "field": 51,
     "not_composed": {}
    },
    "conflict_log": [
     {
      "rule": "D_pres vs channel",
      "result": "no cut: the KO carries every field and archive claim; the compression's bound met with no channel log"
     },
     {
      "rule": "P_coh",
      "result": "F29 (Liu: data 'supporting their outputs … trustworthy and verifiable') against F34/F35 (not 'supporting evidence or citations'), with F23/F24 (a possibility) and F27 (conditional on the user's check) between (sentences 15–17); PE-17 ('only models whose training data is open') against F17/F19 (access to the training data) (sentences 8–9); F45 against F46 (sentence 24): each composed with its source"
     },
     {
      "rule": "N_c",
      "result": "no claim lowered by standing; F23/F24 carried as the hypothesis they are ('may be able to'); F29, F30, F33 at attributed grade; O1075-02 a stipulation of the archive's protocol, composed as one; CAP-1-04 and CAP-1-09 at the registry's interpretation grade"
     },
     {
      "rule": "C_ex",
      "result": "senses composed ⊇ senses in the ledgers: every ledger lineage cited in the KO"
     }
    ]
   }
  },
  "P": {
   "title": "OLMoTrace",
   "genre": "the compression: AIO's interaction grammar (lede, headed clusters, bolded labels, card rail), composed from the v0.9 plan in one claim space; it faces on the card",
   "status": "composed 2026-10-29 from the v0.9 knowledge object; not frozen",
   "lede": {
    "text": "is the Allen Institute for AI's tool, with the University of Washington, UC Berkeley and Stanford, that finds verbatim matches between spans of a language model's output and documents in its full training data, in real time (its paper; census record); its authors state the matched documents are not causes, evidence or citations (the paper).",
    "claims": [
     "F1",
     "F3",
     "F42",
     "PE-01",
     "PE-03",
     "PE-08",
     "F34"
    ]
   },
   "sections": [
    {
     "icon": "🔎",
     "head": "What it carries",
     "items": [
      {
       "label": "Relation:",
       "text": "a verbatim match, one maximal span to the documents containing it, found by lexical overlap (paper; census record).",
       "claims": [
        "F4",
        "PE-04",
        "PE-06",
        "PE-10",
        "F5",
        "F2",
        "PE-21"
       ]
      },
      {
       "label": "Carrier:",
       "text": "the full training corpus on an extended infini-gram index; 3.2 billion documents, 4.6 trillion tokens for OLMo-2-32B-Instruct (paper; Ai2 blog).",
       "claims": [
        "PE-05",
        "F15",
        "F16",
        "PE-07",
        "F9"
       ]
      },
      {
       "label": "Pipeline:",
       "text": "up to 10 snippets per span, BM25-ranked, \"some notion of relevance\"; 4.46 seconds a query (paper; Ai2 blog).",
       "claims": [
        "F10",
        "F11",
        "F12",
        "F13",
        "F14"
       ]
      }
     ]
    },
    {
     "icon": "🧪",
     "head": "Uses and their conditions",
     "items": [
      {
       "label": "Purpose:",
       "text": "where models \"may have learned to generate certain word sequences\" (paper).",
       "claims": [
        "F6",
        "F7",
        "F8"
       ]
      },
      {
       "label": "Uses:",
       "text": "fact checking, hallucination, creativity, math; a false knowledge cutoff traced (paper; Ai2 blog).",
       "claims": [
        "F21",
        "F22",
        "F31",
        "F32",
        "F30",
        "F33"
       ]
      },
      {
       "label": "Fact checking:",
       "text": "users \"may be able to\" check a statement, once they verify the sources are trusted; Liu's launch statement goes further (paper; Ai2 blog; Ken Yeung).",
       "claims": [
        "F23",
        "F24",
        "F25",
        "F26",
        "F27",
        "F28",
        "F29"
       ]
      }
     ]
    },
    {
     "icon": "⚠️",
     "head": "The stated limit",
     "items": [
      {
       "label": "Not causes, evidence or citations:",
       "text": "in both versions of the paper; the census's negative tags (paper; ACL 2025; census record).",
       "claims": [
        "F34",
        "F35",
        "PE-15",
        "PE-16",
        "PE-11"
       ]
      },
      {
       "label": "Associations:",
       "text": "not \"indicators of authority\" (Ai2, to Ken Yeung).",
       "claims": [
        "F36",
        "F37"
       ]
      }
     ]
    },
    {
     "icon": "🧭",
     "head": "Its class",
     "items": [
      {
       "label": "Training/output:",
       "text": "data attribution's relation, where PER measures output/reader (#725).",
       "claims": [
        "O725-01"
       ]
      },
      {
       "label": "Partial, at cost:",
       "text": "recovers no authorship; one post-hoc layer in isolation (#1714; #127).",
       "claims": [
        "O1714-01",
        "O127-01"
       ]
      },
      {
       "label": "Parametric claims:",
       "text": "need training-data provenance records (#95; #1714).",
       "claims": [
        "O95-01",
        "O1714-02"
       ]
      },
      {
       "label": "Evidence:",
       "text": "similarity is exploratory evidence only (#1075).",
       "claims": [
        "O1075-01",
        "O1075-02"
       ]
      },
      {
       "label": "Not RAG:",
       "text": "post hoc, no intervention; Google AI Overview named as RAG (paper).",
       "claims": [
        "F39",
        "F40",
        "F41"
       ]
      }
     ]
    },
    {
     "icon": "🏛️",
     "head": "Origin",
     "items": [
      {
       "label": "Models:",
       "text": "OLMo 2, OLMoE, OLMo 3; any model whose training data the provider holds (Ai2; paper; X).",
       "claims": [
        "F17",
        "F18",
        "F19",
        "F20",
        "F47",
        "PE-17",
        "CAP-1-06",
        "CAP-1-07"
       ]
      },
      {
       "label": "Published:",
       "text": "arXiv April 2025; ACL 2025 Best Demo; open source (arXiv; ACL Anthology).",
       "claims": [
        "F48",
        "F49",
        "PE-13",
        "PE-14",
        "F43",
        "F45",
        "F46",
        "F44",
        "PE-12"
       ]
      },
      {
       "label": "Assessment:",
       "text": "\"generating awareness\"; exposes problematic training content (Ken Yeung; paper).",
       "claims": [
        "F38",
        "F50"
       ]
      }
     ]
    },
    {
     "icon": "📝",
     "head": "Record and composition",
     "items": [
      {
       "label": "At the name:",
       "text": "the Overview offered a match \"as support for a claim\", naming Ai2 alone (Capture Registry, 2026-10-29).",
       "claims": [
        "CAP-1-01",
        "CAP-1-02",
        "CAP-1-10",
        "CAP-1-04",
        "CAP-1-09",
        "CAP-1-08",
        "CAP-1-03",
        "CAP-1-05"
       ]
      },
      {
       "label": "Reception:",
       "text": "no card cites the census record; no Wikipedia article (Capture Registry; Wikipedia).",
       "claims": [
        "CAP-1-11",
        "F51",
        "PE-09",
        "PE-18",
        "PE-19",
        "PE-20"
       ]
      }
     ]
    }
   ],
   "open": [
    {
     "question": "Is fact checking against training data a use the authors offer or a reading their limit excludes?",
     "rests_on": [
      "F21",
      "F22",
      "F23",
      "F24",
      "F25",
      "F27",
      "F29",
      "F34",
      "F35",
      "CAP-1-04"
     ],
     "modality": "self-description and hypothesis (the paper), interpretation (Ai2 blog), attributed (Liu), interpretation (Capture Registry)",
     "would_resolve": "a statement by the authors of what a matched document establishes for a fact-check, beyond 'a more informed judgment'"
    },
    {
     "question": "Does a verbatim match indicate influence, and at what evidentiary grade?",
     "rests_on": [
      "F5",
      "F6",
      "O1075-02",
      "O1714-01",
      "O725-01"
     ],
     "modality": "self-description (the heuristic), stipulation and interpretation (#1075, #1714, #725)",
     "would_resolve": "a validation of the tool's matches against counterfactual retraining or a validated attribution method"
    },
    {
     "question": "Open training data, or access to it?",
     "rests_on": [
      "PE-17",
      "F17",
      "F19",
      "F20",
      "F47"
     ],
     "modality": "stipulation (census) against stipulation and documented (the paper, Ai2 blog), attributed (Liu, Farhadi)",
     "would_resolve": "a deployment of the tool on a model whose training data is held but not public"
    },
    {
     "question": "Fully open-source, or its core?",
     "rests_on": [
      "F45",
      "F46",
      "PE-12"
     ],
     "modality": "documented, both",
     "would_resolve": "the repositories' licences against the deployed system"
    },
    {
     "question": "Which models now, and from which page?",
     "rests_on": [
      "F18",
      "F17",
      "CAP-1-06",
      "CAP-1-07"
     ],
     "modality": "documented (X, snippet only); stipulation (Capture Registry)",
     "would_resolve": "Ai2's own page listing OLMo 3 under the tool, fetched"
    },
    {
     "question": "What do the two unreachable cards say beyond their snippets?",
     "rests_on": [
      "F12",
      "F18"
     ],
     "modality": "documented (snippets only)",
     "would_resolve": "a fetch of the Techzine article and of Ai2's post on X"
    }
   ],
   "rail": [
    {
     "lineage": "what it is: tracing model output to its training data in real time",
     "claims": [
      "PE-01",
      "PE-02",
      "F1",
      "F2"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "the relation: verbatim match between output and training text",
     "claims": [
      "PE-03",
      "F3",
      "F4",
      "PE-04",
      "PE-10",
      "F5"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "what the record's schema asks",
     "claims": [
      "PE-21",
      "PE-09",
      "PE-19",
      "PE-20"
     ],
     "earliest": "PE"
    },
    {
     "lineage": "the pipeline: spans, index, retrieval, ranking",
     "claims": [
      "PE-06",
      "PE-05",
      "F9",
      "F10",
      "F11",
      "F12"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "the evidentiary hierarchy for training-data attribution",
     "claims": [
      "O1075-01",
      "O1075-02"
     ],
     "earliest": 1075
    },
    {
     "lineage": "the training data and its count",
     "claims": [
      "F15",
      "F16",
      "PE-07"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "performance and relevance",
     "claims": [
      "F13",
      "F14"
     ],
     "earliest": "B7"
    },
    {
     "lineage": "models and applicability",
     "claims": [
      "F17",
      "F19",
      "F18",
      "CAP-1-06",
      "CAP-1-07",
      "PE-17",
      "F20"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "openness",
     "claims": [
      "F47",
      "F45",
      "F46",
      "PE-12",
      "F44"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "purpose: where a model may have learned a word sequence",
     "claims": [
      "F6",
      "F7",
      "F8"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "two relations: training/output and output/reader",
     "claims": [
      "O725-01"
     ],
     "earliest": 725
    },
    {
     "lineage": "partial tracing at cost; what survives depends on what is recorded",
     "claims": [
      "O1714-01",
      "O1714-02"
     ],
     "earliest": 1714
    },
    {
     "lineage": "post-hoc detection: one survival layer, developed in isolation",
     "claims": [
      "O127-01"
     ],
     "earliest": 127
    },
    {
     "lineage": "parametric claims and training-data provenance",
     "claims": [
      "O95-01"
     ],
     "earliest": 95
    },
    {
     "lineage": "uses: fact checking, hallucination, creativity, math",
     "claims": [
      "F21",
      "F22",
      "F31",
      "F32",
      "F30"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "fact checking: what a matched document can and cannot show",
     "claims": [
      "F23",
      "F24",
      "F25",
      "F26",
      "F27",
      "F28",
      "F29"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "the stated limit: not causes, not evidence, not citations",
     "claims": [
      "F34",
      "F35",
      "PE-15",
      "PE-16",
      "PE-11",
      "F36",
      "F37",
      "CAP-1-03"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "uses: regulation and auditing",
     "claims": [
      "F33"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "unlike retrieval-augmented generation",
     "claims": [
      "F39",
      "F40",
      "F41"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "the composition at the name: what it held",
     "claims": [
      "CAP-1-01",
      "CAP-1-02",
      "CAP-1-10"
     ],
     "earliest": "CAP"
    },
    {
     "lineage": "the composition converts a match into support",
     "claims": [
      "CAP-1-04",
      "CAP-1-09",
      "CAP-1-08"
     ],
     "earliest": "CAP"
    },
    {
     "lineage": "the operator and the authors",
     "claims": [
      "F43",
      "F42",
      "PE-08",
      "CAP-1-05"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "publication",
     "claims": [
      "F48",
      "PE-14",
      "F49",
      "PE-13"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "risks",
     "claims": [
      "F50"
     ],
     "earliest": "B7"
    },
    {
     "lineage": "assessment from outside",
     "claims": [
      "F38"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "neighbours in the census",
     "claims": [
      "PE-18"
     ],
     "earliest": "PE"
    },
    {
     "lineage": "reception: the record and the cards",
     "claims": [
      "CAP-1-11"
     ],
     "earliest": "CAP"
    },
    {
     "lineage": "reception: the encyclopedia",
     "claims": [
      "F51"
     ],
     "earliest": "B8"
    }
   ],
   "body_words": 323,
   "entity": "Composed under v0.9, equal case (ρ = extends): KO_B sentences surviving 0 of 14; every field claim carried."
  },
  "P_B": {
   "title": "OLMoTrace",
   "arm": "B (the field alone)",
   "genre": "popup — the compression from the field ledger alone (cards, own page, Wikipedia)",
   "status": "composed 2026-10-29 from the field ledger alone; not frozen",
   "lede": {
    "text": "is the Allen Institute for AI's system that traces a language model's output back to its full, multi-trillion-token training data in real time, by verbatim match (its paper; Ai2 blog).",
    "claims": [
     "F1",
     "F2",
     "F3",
     "F42"
    ]
   },
   "sections": [
    {
     "icon": "🔎",
     "head": "How it works",
     "items": [
      {
       "label": "Match:",
       "text": "long, unique spans found verbatim in the training data; retrieval by lexical overlap (Ai2 blog; paper).",
       "claims": [
        "F4",
        "F5"
       ]
      },
      {
       "label": "Index:",
       "text": "infini-gram, O(L * log N); up to 10 snippets per span, BM25-ranked (Ai2 blog; paper).",
       "claims": [
        "F9",
        "F10",
        "F11",
        "F12",
        "F13",
        "F14"
       ]
      },
      {
       "label": "Data:",
       "text": "3.2 billion documents, 4.6 trillion tokens for OLMo-2-32B-Instruct (paper).",
       "claims": [
        "F15",
        "F16"
       ]
      }
     ]
    },
    {
     "icon": "🧪",
     "head": "Uses",
     "items": [
      {
       "label": "Purpose:",
       "text": "where models \"may have learned\" word sequences (paper).",
       "claims": [
        "F6",
        "F7",
        "F8"
       ]
      },
      {
       "label": "Fact checking:",
       "text": "a possibility, once users verify the sources are trusted (paper; Ai2 blog; Ken Yeung).",
       "claims": [
        "F21",
        "F22",
        "F23",
        "F24",
        "F25",
        "F26",
        "F27",
        "F28",
        "F29",
        "F30",
        "F31",
        "F32",
        "F33"
       ]
      }
     ]
    },
    {
     "icon": "⚠️",
     "head": "Limit",
     "items": [
      {
       "label": "Not causes, evidence or citations:",
       "text": "the documents, in the authors' words (paper; ACL 2025; Ken Yeung).",
       "claims": [
        "F34",
        "F35",
        "F36",
        "F37"
       ]
      },
      {
       "label": "Not RAG:",
       "text": "post hoc, no intervention (paper).",
       "claims": [
        "F39",
        "F40",
        "F41"
       ]
      }
     ]
    },
    {
     "icon": "🏛️",
     "head": "Origin",
     "items": [
      {
       "label": "Models and access:",
       "text": "OLMo 2, OLMoE, OLMo 3; any model with its training data available (Ai2 blog; paper; X).",
       "claims": [
        "F17",
        "F18",
        "F19",
        "F20"
       ]
      },
      {
       "label": "Published:",
       "text": "31 authors; open source; ACL 2025 Best Demo; no Wikipedia article (arXiv; ACL Anthology; Wikipedia).",
       "claims": [
        "F43",
        "F44",
        "F45",
        "F46",
        "F47",
        "F48",
        "F49",
        "F50",
        "F38",
        "F51"
       ]
      }
     ]
    }
   ],
   "open": [
    {
     "question": "Is fact checking a use or a reading the limit excludes?",
     "rests_on": [
      "F23",
      "F27",
      "F29",
      "F34"
     ],
     "modality": "hypothesis, interpretation, attributed, self-description",
     "would_resolve": "the authors' statement of what a matched document establishes"
    },
    {
     "question": "Fully open-source, or its core?",
     "rests_on": [
      "F45",
      "F46"
     ],
     "modality": "documented, both",
     "would_resolve": "the repositories' licences"
    }
   ],
   "rail": [
    {
     "lineage": "what it is: tracing model output to its training data in real time",
     "claims": [
      "F1",
      "F2"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "the relation: verbatim match between output and training text",
     "claims": [
      "F3",
      "F4",
      "F5"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "the training data and its count",
     "claims": [
      "F15",
      "F16"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "the pipeline: spans, index, retrieval, ranking",
     "claims": [
      "F9",
      "F10",
      "F11",
      "F12"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "performance and relevance",
     "claims": [
      "F13",
      "F14"
     ],
     "earliest": "B7"
    },
    {
     "lineage": "models and applicability",
     "claims": [
      "F17",
      "F19",
      "F18",
      "F20"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "purpose: where a model may have learned a word sequence",
     "claims": [
      "F6",
      "F7",
      "F8"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "uses: fact checking, hallucination, creativity, math",
     "claims": [
      "F21",
      "F22",
      "F31",
      "F32",
      "F30"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "uses: regulation and auditing",
     "claims": [
      "F33"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "fact checking: what a matched document can and cannot show",
     "claims": [
      "F23",
      "F24",
      "F25",
      "F26",
      "F27",
      "F28",
      "F29"
     ],
     "earliest": "B1"
    },
    {
     "lineage": "the stated limit: not causes, not evidence, not citations",
     "claims": [
      "F34",
      "F35",
      "F36",
      "F37"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "unlike retrieval-augmented generation",
     "claims": [
      "F39",
      "F40",
      "F41"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "the operator and the authors",
     "claims": [
      "F43",
      "F42"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "openness",
     "claims": [
      "F45",
      "F46",
      "F44",
      "F47"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "publication",
     "claims": [
      "F48",
      "F49"
     ],
     "earliest": "B2"
    },
    {
     "lineage": "risks",
     "claims": [
      "F50"
     ],
     "earliest": "B7"
    },
    {
     "lineage": "assessment from outside",
     "claims": [
      "F38"
     ],
     "earliest": "B6"
    },
    {
     "lineage": "reception: the encyclopedia",
     "claims": [
      "F51"
     ],
     "earliest": "B8"
    }
   ],
   "body_words": 158,
   "entity": "With no archive claim admitted, the entry is Ai2's tracing tool at its authors' resolution: its paper in three instances, its launch post, the launch coverage, and its stated limit."
  },
  "KO_B": {
   "title": "OLMoTrace",
   "arm": "B (the field alone)",
   "genre": "knowledge object",
   "status": "composed 2026-10-29; not frozen",
   "sentences": [
    {
     "n": 1,
     "para": 0,
     "text": "OLMoTrace is, in its authors' words, \"the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time\" (arXiv:2504.07096), introduced by Ai2 on 9 April 2025 as a feature of the Ai2 Playground (Ai2 blog).",
     "claims": [
      "F1",
      "F2"
     ]
    },
    {
     "n": 2,
     "para": 0,
     "text": "It finds and shows verbatim matches between segments of a model's output and documents in the training corpora, highlighting long and unique spans that occur verbatim at least once in the training data (arXiv abstract; Ai2 blog), and it retrieves by lexical overlap on the heuristic that such documents are likely to be influential (the paper, §6).",
     "claims": [
      "F3",
      "F4",
      "F5"
     ]
    },
    {
     "n": 3,
     "para": 1,
     "text": "It matches against the whole of a model's training data, for OLMo-2-32B-Instruct 3.2 billion documents and 4.6 trillion tokens (the paper; Ai2 blog).",
     "claims": [
      "F15",
      "F16"
     ]
    },
    {
     "n": 4,
     "para": 1,
     "text": "Spans are found over an infini-gram index at O(L * log N) (Ai2 blog), up to 10 enclosing snippets are retrieved per kept span and the documents ranked by BM25, a score that conveys \"some notion of relevance\" (the paper; Ai2 blog), and the user starts it with the \"Show OLMoTrace\" button below an answer (Techzine, snippet only).",
     "claims": [
      "F9",
      "F10",
      "F11",
      "F12"
     ]
    },
    {
     "n": 5,
     "para": 1,
     "text": "The paper measures 4.46 seconds per query and a first displayed document of 1.90 on a 0–3 relevance scale, roughly \"on the same topic\" (the paper).",
     "claims": [
      "F13",
      "F14"
     ]
    },
    {
     "n": 6,
     "para": 2,
     "text": "At launch it served OLMo 2 32B and 13B Instruct and OLMoE 1B 7B Instruct and can be applied to any model whose full training data the provider has access to (Ai2 blog; the paper); Ai2's post on X gives the OLMo 3 family (X, snippet only), and Liu expects closed-model providers to hesitate (Ken Yeung).",
     "claims": [
      "F17",
      "F19",
      "F18",
      "F20"
     ]
    },
    {
     "n": 7,
     "para": 3,
     "text": "Its stated purpose is to explore where models \"may have learned to generate certain word sequences\", verbatim matching taken as the most direct connection (the paper; Ai2 blog), at an early stage of exploring the tool (Ai2 blog).",
     "claims": [
      "F6",
      "F7",
      "F8"
     ]
    },
    {
     "n": 8,
     "para": 4,
     "text": "Its authors show it on fact checking, hallucination and creativity, with case studies on fact checking, \"creative\" expressions and math (the paper), a memorized math expression and a false knowledge cutoff traced to post-training data (Ai2 blog), and Ai2's leaders tie it to hallucination scores and to auditability (Ken Yeung).",
     "claims": [
      "F21",
      "F22",
      "F31",
      "F32",
      "F30",
      "F33"
     ]
    },
    {
     "n": 9,
     "para": 4,
     "text": "Fact checking is stated as a possibility, users \"may be able to fact-check the statement against its training data\", where context and source help a more informed judgment because some sources are unreliable (the paper, both versions), and confidence follows only if users personally verify that the documents come from trusted sources (Ai2 blog); the newsletter calls it fact-checking, and Liu's launch statement says users can ensure the data supporting their outputs is \"trustworthy and verifiable\" (Ken Yeung).",
     "claims": [
      "F23",
      "F24",
      "F25",
      "F26",
      "F27",
      "F28",
      "F29"
     ]
    },
    {
     "n": 10,
     "para": 5,
     "text": "The authors' limit: \"The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output\" (the paper, arXiv v2 and ACL 2025); Ai2 told the newsletter the documents are associations, not the definitive source and not indicators of authority (Ken Yeung).",
     "claims": [
      "F34",
      "F35",
      "F36",
      "F37"
     ]
    },
    {
     "n": 11,
     "para": 6,
     "text": "The paper sets it apart from retrieval-augmented generation, Google AI Overview among RAG's examples: it retrieves post hoc and does not intervene in generation (the paper; Liu in Ken Yeung), and Gemini's double-check is less useful because Search is not Gemini's training data (the paper).",
     "claims": [
      "F39",
      "F40",
      "F41"
     ]
    },
    {
     "n": 12,
     "para": 7,
     "text": "Its 31 authors are at the Allen Institute for AI, the University of Washington, UC Berkeley and Stanford (arXiv; the paper); it is \"publicly available and fully open-source\", its core under Apache 2.0, the openness Ai2 calls its commitment and Farhadi its condition (arXiv; the paper; Ai2 blog; Ken Yeung).",
     "claims": [
      "F43",
      "F42",
      "F45",
      "F46",
      "F44",
      "F47"
     ]
    },
    {
     "n": 13,
     "para": 7,
     "text": "Submitted 9 April 2025 and revised 8 July, it appeared in ACL 2025's System Demonstrations with the Best Demo award (arXiv; ACL Anthology); its authors name the easier exposure of problematic training content as a risk (the paper), and the newsletter finds it gives no analytics for improving data, \"generating awareness\" (Ken Yeung).",
     "claims": [
      "F48",
      "F49",
      "F50",
      "F38"
     ]
    },
    {
     "n": 14,
     "para": 8,
     "text": "Wikipedia has no article on it (Wikipedia, 2026-10-29).",
     "claims": [
      "F51"
     ]
    }
   ]
  },
  "L_B": {
   "words": 648,
   "text": "OLMoTrace is, in its authors' words, \"the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time\" (arXiv:2504.07096), introduced by Ai2 on 9 April 2025 as a feature of the Ai2 Playground (Ai2 blog). It finds and shows verbatim matches between segments of a model's output and documents in the training corpora, highlighting long and unique spans that occur verbatim at least once in the training data (arXiv abstract; Ai2 blog), and it retrieves by lexical overlap on the heuristic that such documents are likely to be influential (the paper, §6). It matches against the whole of a model's training data, for OLMo-2-32B-Instruct 3.2 billion documents and 4.6 trillion tokens (the paper; Ai2 blog). Spans are found over an infini-gram index at O(L * log N) (Ai2 blog), up to 10 enclosing snippets are retrieved per kept span and the documents ranked by BM25, a score that conveys \"some notion of relevance\" (the paper; Ai2 blog), and the user starts it with the \"Show OLMoTrace\" button below an answer (Techzine, snippet only). The paper measures 4.46 seconds per query and a first displayed document of 1.90 on a 0–3 relevance scale, roughly \"on the same topic\" (the paper). At launch it served OLMo 2 32B and 13B Instruct and OLMoE 1B 7B Instruct and can be applied to any model whose full training data the provider has access to (Ai2 blog; the paper); Ai2's post on X gives the OLMo 3 family (X, snippet only), and Liu expects closed-model providers to hesitate (Ken Yeung). Its stated purpose is to explore where models \"may have learned to generate certain word sequences\", verbatim matching taken as the most direct connection (the paper; Ai2 blog), at an early stage of exploring the tool (Ai2 blog). Its authors show it on fact checking, hallucination and creativity, with case studies on fact checking, \"creative\" expressions and math (the paper), a memorized math expression and a false knowledge cutoff traced to post-training data (Ai2 blog), and Ai2's leaders tie it to hallucination scores and to auditability (Ken Yeung). Fact checking is stated as a possibility, users \"may be able to fact-check the statement against its training data\", where context and source help a more informed judgment because some sources are unreliable (the paper, both versions), and confidence follows only if users personally verify that the documents come from trusted sources (Ai2 blog); the newsletter calls it fact-checking, and Liu's launch statement says users can ensure the data supporting their outputs is \"trustworthy and verifiable\" (Ken Yeung). The authors' limit: \"The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output\" (the paper, arXiv v2 and ACL 2025); Ai2 told the newsletter the documents are associations, not the definitive source and not indicators of authority (Ken Yeung). The paper sets it apart from retrieval-augmented generation, Google AI Overview among RAG's examples: it retrieves post hoc and does not intervene in generation (the paper; Liu in Ken Yeung), and Gemini's double-check is less useful because Search is not Gemini's training data (the paper). Its 31 authors are at the Allen Institute for AI, the University of Washington, UC Berkeley and Stanford (arXiv; the paper); it is \"publicly available and fully open-source\", its core under Apache 2.0, the openness Ai2 calls its commitment and Farhadi its condition (arXiv; the paper; Ai2 blog; Ken Yeung). Submitted 9 April 2025 and revised 8 July, it appeared in ACL 2025's System Demonstrations with the Best Demo award (arXiv; ACL Anthology); its authors name the easier exposure of problematic training content as a risk (the paper), and the newsletter finds it gives no analytics for improving data, \"generating awareness\" (Ken Yeung). Wikipedia has no article on it (Wikipedia, 2026-10-29).",
   "note": "L_F(B), the field composed in the field's ontology (v0.9 §0.4), realized as KO_B; its rail is P_B's, by lineage."
  },
  "L_BA": {
   "words": 1388,
   "text": "OLMoTrace is a tracing tool in the census's lineage of training data, attaching provenance to \"a span of a language model's output and the training documents that contain it verbatim\" (provenance census record), and its authors present it as \"the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time\" (arXiv:2504.07096). Ai2 introduced it on 9 April 2025 as a feature of the Ai2 Playground (Ai2 blog), and the census answers its organizing question for it, \"What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?\", field by field (census record). The relation it carries is a verbatim match: it \"finds and shows verbatim matches between segments of language model output and documents in the training text corpora\" (arXiv abstract), highlighting long and unique spans found at least once in the training data (Ai2 blog), which the census records as its one relation type, its operation (model generation traced back to training text) and its unit, \"a maximal matching span of model output\" (census record). It retrieves \"by lexical overlap, with the heuristic that such training examples are likely to be influential for the given output\" (the paper, §6); for the class of training-data attribution, which treats model behaviour as counterfactually dependent on the examples that entered training, the archive's protocol ranks exact paired retraining first, validated approximation second, and \"unvalidated similarity or intuition only as exploratory evidence\" (#1075). Its carrier is the model's full training corpus over an extended infini-gram index (census record): for OLMo-2-32B-Instruct, 3.2 billion documents and 4.6 trillion tokens across pre-, mid- and post-training, all of them matched (the paper; Ai2 blog; census record). Span finding runs at O(L * log N) (Ai2 blog); up to 10 enclosing snippets are retrieved per kept span and ranked by BM25 (the paper), a score that conveys \"some notion of relevance\" (Ai2 blog), behind a \"Show OLMoTrace\" button below the model's answer (Techzine, snippet only). Per query the tracing takes 4.46 seconds on average, and the first document it displays averages 1.90 on a 0–3 relevance scale, roughly \"on the same topic\" as the output, in the paper's own measurement (the paper). At launch it served OLMo 2 32B and 13B Instruct and OLMoE 1B 7B Instruct and \"can be applied to any language model for which you have access to the training data\" (Ai2 blog), \"as long as the service provider has access to its full training data\" (the paper); Ai2's post on X gives the OLMo 3 family (X, snippet only), which the Capture Registry finds the census record does not yet carry and is to carry from Ai2's own page (Capture Registry, 2026-10-29). The census tags it as tracing \"only models whose training data is open\" (census record); Farhadi ties the connection to Ai2's open inputs, and Liu expects closed-model providers to hesitate because the tool would expose their training data (Ken Yeung). Its purpose, in the paper, is to explore where models \"may have learned to generate certain word sequences\", verbatim matching taken as \"the most direct connection between LM outputs and the training data\" (the paper), for researchers, developers and the general public (Ai2 blog), at \"an early stage\" of exploring the tool (Ai2 blog). The archive places the class this tool belongs to on the training/output relation: data attribution asks \"Which training examples shaped a specific output\", an adjacent method that \"Operates on training/output relation; PER operates on output/reader relation\" (#725). Of that class the archive holds that partial tracing \"is possible at cost\", that it does not recover authorship, and that \"what survives depends on what is recorded and computed\" (#1714), and that training-data attribution is the post-hoc detection layer, developed \"in isolation\" from the other survival layers, no system bundling them (#127). Claims drawn from a model's training-time parameters, on the archive's typology, often carry no surface signal of the sources that contributed; PER reaches them only indirectly, and the substrate level needs companion infrastructure, \"training-data provenance records\" among it (#95), since provenance \"has to survive the training pipeline\" (#1714). Its authors show it on \"fact checking, hallucination, and the creativity of language models\" (arXiv abstract), with three case studies, fact checking first (the paper), and Ai2's post traces a math expression the model could have learned by memorization and a false knowledge cutoff to post-training examples, a finding that changed the 32B model's post-training data (Ai2 blog). Fact checking stands in the paper as a possibility, users \"may be able to fact-check the statement against its training data\", with its qualification that context and source help \"a more informed judgment on the factuality of the statement\", since words mislead out of context and some web sources are unreliable (the paper, arXiv v2 and ACL 2025). Ai2's post calls the interpretation of matches \"quite nuanced\" and makes confidence in a generated fact conditional on the user's own check that the documents \"come from trusted sources\" (Ai2 blog); the newsletter calls it an application \"offering fact-checking\", and Liu's launch statement says users \"can ensure that the data supporting their outputs is trustworthy and verifiable\" (Ken Yeung). The authors state the limit in both versions of the paper, \"The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output\" (the paper; ACL 2025), and the census carries it as two negative tags, not causes of the output and \"NOT a citation system\", and in what it records as surviving (census record). To the newsletter Ai2 called the documents associations, not \"the definitive source of the model's learning\", and its results not \"indicators of authority\" (Ken Yeung). Ai2's leaders tie the tool to hallucination scores, \"assuming that the inputs were factual\" (Farhadi), and to regulation, governance and auditability (Lebrecht), in the newsletter's report (Ken Yeung). The paper sets it apart from retrieval-augmented generation, naming Google AI Overview among RAG's examples: OLMoTrace \"retrieves documents post-hoc and does not intervene with the LM generation\" (the paper; Liu in Ken Yeung), and Gemini's double-check, drawn from Search and not from Gemini's training data, is \"less useful for scientific exploration\" (the paper). At its own name on Google's AI Overview, 2026-10-29, the composition held the tool, its maker, its real time, its scale and its verbatim matching, checked against the census record and the pages it saved, naming the tool, its method, its models and its uses (Capture Registry, 2026-10-29), and gave as its first use \"Fact-Checking: Verify whether specific claims in a model's answer are backed by real training documents\", a verbatim match \"offered as support for a claim, the reading the paper excludes\" (Capture Registry, 2026-10-29). The registry reads this as \"the conversion of a matching relation into a citation relation that the paper names and refuses\", a tool whose authors state what a match does not mean \"composed with that meaning restored\", and it finds the census record's negative tags to be the authors' own (Capture Registry, 2026-10-29). Its 31 authors are at the Allen Institute for AI, the University of Washington, UC Berkeley and Stanford (arXiv; the paper), the operator the census records whole (census record), where the composition named Ai2 alone (Capture Registry, 2026-10-29). It is \"publicly available and fully open-source\" (arXiv abstract), its core under the Apache 2.0 license (the paper), reached through the Ai2 Playground (census record), an expression of Ai2's commitment to \"open models, open data, and beyond\" (Ai2 blog). Submitted 9 April 2025 and revised 8 July (arXiv; census record), it appeared in ACL 2025's System Demonstrations with the Best Demo award (ACL Anthology), and the census records it active, \"ACL 2025 demo track\" (census record). Its authors name as a risk that it makes problematic training content \"more easily exposed\" (the paper), and the newsletter finds that it gives no analytics for improving the data: \"Simply put, it's generating awareness\" (Ken Yeung). The census records the tool as one that records provenance, a neighbour of the Data Provenance Initiative, with the Playground and \"infini-gram as a separate record\" left unchecked (census record); no card cites the record, which landed the same day (Capture Registry, 2026-10-29), and Wikipedia has no article on OLMoTrace (Wikipedia, 2026-10-29).",
   "rail": [],
   "note": "L(B ∪ A) in the equal-terms case (v0.9 §0.4), realized as the knowledge object (objects.KO); its rail is the compression's, by lineage."
  }
 },
 "field": [
  {
   "id": "B1",
   "card": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
   "fetched": "card; B1.html sha256 39e59dda3ba78e0b, HTTP 200 at 2026-10-29T17:07:13Z",
   "origin": "card"
  },
  {
   "id": "B2",
   "card": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
   "fetched": "card; B2.html sha256 cd982680a1ca4c02, HTTP 200 at 2026-10-29T17:07:17Z; byte-identical to the census copy olmotrace--abs",
   "origin": "card"
  },
  {
   "id": "B3",
   "card": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
   "fetched": "card; B3.html sha256 71292b784f0dc598, HTTP 200 at 2026-10-29T17:07:19Z; PDF sha256 fcac45353235b662",
   "origin": "card"
  },
  {
   "id": "B4",
   "card": "X — Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
   "fetched": "unreachable: https://x.com/allen_ai HTTP 200; https://x.com/search?q=%22wired%20into%20OlmoTrace%22 HTTP 200; card title and snippet only",
   "origin": "card"
  },
  {
   "id": "B5",
   "card": "Techzine Global — Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
   "fetched": "unreachable: https://www.techzine.eu/news/analytics/130296/ai2-introduces-olmotrace-a-peek-inside-llms/ HTTP 403; https://www.techzine.eu/news/analytics/130296/ai2-introduces-olmotrace-a-peek-inside-llms/ HTTP 403; https://www.techzine.eu/news/analytics/130296/ai2-introduces-olmotrace-a-peek-inside-llms/ HTTP 403; card title and snippet only",
   "origin": "card"
  },
  {
   "id": "B6",
   "card": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
   "fetched": "card; B6.html sha256 bf021478cec54cb8, HTTP 200 at 2026-10-29T17:07:23Z",
   "origin": "card"
  },
  {
   "id": "B7",
   "card": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
   "fetched": "own_page; census copy olmotrace--paper (read 2026-10-29, HTTP 200); B7.html sha256 b1b16ba583eeb0db",
   "origin": "own_page"
  },
  {
   "id": "B8",
   "card": "Wikipedia — OLMoTrace (no article)",
   "fetched": "no article: raw HTTP 404, rendered HTTP 404 (after one 429), search no results; sha256 92d530a8e9a8411f",
   "origin": "wikipedia"
  }
 ],
 "field_claims": {
  "F1": {
   "source": "B2",
   "origin": "card",
   "locus": "Abstract s1",
   "quote": "We present OLMoTrace, the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time.",
   "claim": "OLMoTrace defined by its authors: the first system to trace language-model outputs back to their full, multi-trillion-token training data in real time",
   "modality": "stipulation",
   "kind": "definition",
   "kernel": true,
   "const": true,
   "lineage": "what it is: tracing model output to its training data in real time",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F2": {
   "source": "B1",
   "origin": "card",
   "locus": "¶1",
   "quote": "Today we introduce OLMoTrace, a one-of-a-kind feature in the Ai2 Playground that lets you trace the outputs of language models back to their full, multi-trillion-token training data in real time.",
   "claim": "the operator's post introduces it as a feature of the Ai2 Playground that traces model outputs to their full training data in real time",
   "modality": "stipulation",
   "kind": "definition",
   "kernel": true,
   "const": true,
   "lineage": "what it is: tracing model output to its training data in real time",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F3": {
   "source": "B2",
   "origin": "card",
   "locus": "Abstract s2",
   "quote": "OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora.",
   "claim": "what it finds: verbatim matches between segments of model output and documents in the training corpora",
   "modality": "stipulation",
   "kind": "definition",
   "kernel": true,
   "const": true,
   "lineage": "the relation: verbatim match between output and training text",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F4": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interact with OLMoTrace' ¶1",
   "quote": "The highlights indicate long and unique spans that appear verbatim at least once in the training data of this model.",
   "claim": "the highlighted spans are long and unique spans that occur verbatim at least once in the model's training data",
   "modality": "stipulation",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "the relation: verbatim match between output and training text",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F5": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§6 Related Work, 'Tracing LM generation into training data'",
   "quote": "we directly retrieve similar training examples by lexical overlap, with the heuristic that such training examples are likely to be influential for the given output.",
   "claim": "its method against influence functions: retrieval by lexical overlap, on the heuristic that such examples are likely to be influential",
   "modality": "self-description",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "the relation: verbatim match between output and training text",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F6": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§1 ¶4",
   "quote": "The purpose of OLMoTrace is to give users a tool to explore where LMs may have learned to generate certain word sequences, focusing on verbatim matching as the most direct connection between LM outputs and the training data.",
   "claim": "its purpose: to explore where models may have learned to generate word sequences, verbatim matching taken as the most direct connection",
   "modality": "stipulation",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "purpose: where a model may have learned a word sequence",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F7": {
   "source": "B1",
   "origin": "card",
   "locus": "¶2",
   "quote": "We developed OLMoTrace to enable researchers, developers, and the general public to inspect where and how language models may have learned to generate certain word sequences.",
   "claim": "developed so that researchers, developers and the public can inspect where and how models may have learned word sequences",
   "modality": "stipulation",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "purpose: where a model may have learned a word sequence",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F8": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interpreting results' ¶5",
   "quote": "We are still in an early stage in exploring the full potential of this tool",
   "claim": "the operator places the tool at an early stage of exploring its potential",
   "modality": "self-description",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "purpose: where a model may have learned a word sequence",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F9": {
   "source": "B1",
   "origin": "card",
   "locus": "'OLMoTrace behind the scenes' ¶3",
   "quote": "reducing the time complexity of span finding from the naive O(L^2 * N) to O(L * log N)",
   "claim": "span finding over an infini-gram index, its time complexity reduced from O(L^2 * N) to O(L * log N)",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "the pipeline: spans, index, retrieval, ranking",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F10": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§3 Steps 3 and 5",
   "quote": "For each kept span, we retrieve up to 10 document snippets from the training data that enclose this span. … we rank all documents by a BM25 score in descending order.",
   "claim": "up to 10 enclosing document snippets retrieved per kept span; the documents ranked by BM25",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "the pipeline: spans, index, retrieval, ranking",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F11": {
   "source": "B1",
   "origin": "card",
   "locus": "'OLMoTrace behind the scenes' ¶7",
   "quote": "We find that this BM25 score conveys some notion of “relevance”, and thus we surface documents with higher scores on top.",
   "claim": "the BM25 ranking conveys 'some notion of relevance'",
   "modality": "self-description",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "the pipeline: spans, index, retrieval, ranking",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F12": {
   "source": "B5",
   "origin": "card",
   "locus": "card snippet (Techzine unreachable: HTTP 403)",
   "quote": "After generating an answer from OLMo in the Ai2 Playground, users can activate the tool by clicking the “Show OLMoTrace” button below the output.",
   "claim": "activated in the Playground by the 'Show OLMoTrace' button below a model's answer",
   "modality": "documented",
   "kind": "fact",
   "kernel": true,
   "const": false,
   "lineage": "the pipeline: spans, index, retrieval, ranking",
   "date": "undated on the card",
   "quote_ok": true,
   "checked_against": "card title and snippet (unreachable card)"
  },
  "F13": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§3.2",
   "quote": "the OLMoTrace inference latency per query is 4.46 seconds",
   "claim": "measured latency: 4.46 seconds per query on 98 internal conversations",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "performance and relevance",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F14": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§4 'Validating the relevance rankings'",
   "quote": "The first document displayed in each conversation received an average relevance score of 1.90 (roughly meaning “being on the same topic as the LM output)",
   "claim": "the first displayed document scored 1.90 of 3 on human-rated relevance, roughly 'on the same topic'",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "performance and relevance",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F15": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§2 'The training data.'",
   "quote": "which totals 3.2 billion documents and 4.6 trillion (Llama-2) tokens",
   "claim": "OLMo-2-32B-Instruct's training data, matched in full: 3.2 billion documents, 4.6 trillion tokens",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "the training data and its count",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F16": {
   "source": "B1",
   "origin": "card",
   "locus": "'OLMoTrace behind the scenes' ¶8",
   "quote": "In total, these five datasets contain about 3.2 billion documents and 4.6 trillion tokens, and OLMoTrace matches against all of them.",
   "claim": "the five datasets of pre-, mid- and post-training, about 3.2 billion documents and 4.6 trillion tokens, all matched",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "the training data and its count",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F17": {
   "source": "B1",
   "origin": "card",
   "locus": "¶2",
   "quote": "OLMoTrace is available today with our three flagship models: OLMo 2 32B Instruct, OLMo 2 13B Instruct, and OLMoE 1B 7B Instruct, and it can be applied to any language model for which you have access to the training data.",
   "claim": "at launch three models (OLMo 2 32B and 13B Instruct, OLMoE 1B 7B Instruct); applicable to any model whose training data one has access to",
   "modality": "documented",
   "kind": "fact",
   "kernel": true,
   "const": false,
   "lineage": "models and applicability",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F18": {
   "source": "B4",
   "origin": "card",
   "locus": "card title and snippet (X unreachable)",
   "quote": "Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just see its behavior—you can trace it. For example, you can ask Olmo 3-Think (32B",
   "claim": "Ai2 on X: OLMo 3 traced in the Playground, OLMo 3-Think (32B) among the models",
   "modality": "documented",
   "kind": "fact",
   "kernel": true,
   "const": false,
   "lineage": "models and applicability",
   "date": "undated on the card (OLMo 3; after 2025-11)",
   "quote_ok": true,
   "checked_against": "card title and snippet (unreachable card)"
  },
  "F19": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§1 ¶5",
   "quote": "OLMoTrace can be applied to any LM as long as the service provider has access to its full training data.",
   "claim": "applicable to any language model, on the condition that the service provider has access to its full training data",
   "modality": "stipulation",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "models and applicability",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F20": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'OLMoTrace isn't limited to Ai2's model'",
   "quote": "However, Liu reasons that closed model providers may hesitate to use OLMoTrace because it would expose the training data to sunlight.",
   "claim": "Liu, reported: closed-model providers may hesitate to use it because it would expose their training data",
   "modality": "attributed",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "models and applicability",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F21": {
   "source": "B2",
   "origin": "card",
   "locus": "Abstract s5",
   "quote": "We showcase how it can be used to explore fact checking, hallucination, and the creativity of language models.",
   "claim": "uses shown by the authors: exploring fact checking, hallucination and creativity",
   "modality": "self-description",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "uses: fact checking, hallucination, creativity, math",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F22": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§1 ¶4",
   "quote": "We present three case studies for ways to use OLMoTrace (§ 5 ): (1) fact checking, (2) tracing the LM-generated “creative” expressions, and (3) tracing math capabilities.",
   "claim": "three case studies, fact checking first, then 'creative' expressions and math capabilities",
   "modality": "self-description",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "uses: fact checking, hallucination, creativity, math",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F23": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§5 'Fact checking.'",
   "quote": "If the LM states a fact, users may be able to fact-check the statement against its training data.",
   "claim": "fact checking stated as a possibility: users may be able to fact-check a model's statement against its training data",
   "modality": "hypothesis",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F24": {
   "source": "B3",
   "origin": "card",
   "locus": "PDF §5 'Fact checking.'",
   "quote": "If the LM states a fact, users may be able to fact-check the statement against its training data.",
   "claim": "the published version carries the same conditional",
   "modality": "hypothesis",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-07 (ACL 2025 proceedings)",
   "quote_ok": true,
   "checked_against": "saved text B3.txt (sha256 26daccd5e52cd634…), B3-pdf.txt (sha256 f2fc53f52892169d…), whitespace-normalised"
  },
  "F25": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§5 'Fact checking.' ¶2",
   "quote": "We note that inspecting the document context and source can help users make a more informed judgment on the factuality of the statement, as words can be misleading out of context, and some web sources may be unreliable.",
   "claim": "its qualification: the document's context and source help a more informed judgment, since words mislead out of context and some sources are unreliable",
   "modality": "self-description",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F26": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interpreting results' ¶1",
   "quote": "The interpretation of text matches with training data is quite nuanced. For some highlighted spans in the model response, we can use their corresponding documents to fact check claims made in the spans.",
   "claim": "the operator's post: interpreting matches is 'quite nuanced'; for some spans the documents can be used to fact-check the span's claims",
   "modality": "self-description",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F27": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interpreting results' ¶2",
   "quote": "If you personally verify that these documents come from trusted sources, you can gain confidence that this piece of information generated by the model is factually correct.",
   "claim": "confidence in a generated fact is conditional on the user personally verifying that the matched documents come from trusted sources",
   "modality": "interpretation",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F28": {
   "source": "B6",
   "origin": "card",
   "locus": "¶1",
   "quote": "Called OLMoTrace, it's an open-source application offering fact-checking of information provided in prompt replies.",
   "claim": "the newsletter: an open-source application offering fact-checking of replies",
   "modality": "interpretation",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F29": {
   "source": "B6",
   "origin": "card",
   "locus": "¶2 (Liu, in a statement)",
   "quote": "anyone using our models can ensure that the data supporting their outputs is trustworthy and verifiable.",
   "claim": "Liu, in Ai2's launch statement: users can ensure that the data 'supporting their outputs' is trustworthy and verifiable",
   "modality": "attributed",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "fact checking: what a matched document can and cannot show",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F30": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'OLMoTrace connects the output to the input' (Farhadi)",
   "quote": "And the minute you start linking the generation to an input—assuming that the inputs were factual—then you can start also cooking up scores for hallucination",
   "claim": "Farhadi, reported: once generation is linked to input, and assuming the inputs factual, hallucination scores become possible",
   "modality": "attributed",
   "kind": "prediction",
   "kernel": false,
   "const": false,
   "lineage": "uses: fact checking, hallucination, creativity, math",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F31": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interpreting results' ¶4",
   "quote": "OLMoTrace reveals that this entire expression appears multiple times in the training data, so the model could have easily learned it through memorization.",
   "claim": "a math expression found many times in the training data: the model could have learned it by memorization",
   "modality": "interpretation",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "uses: fact checking, hallucination, creativity, math",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F32": {
   "source": "B1",
   "origin": "card",
   "locus": "'Interpreting results' ¶5",
   "quote": "now we can see that the model may have learned to say August 2023 from the post-training examples listed on the right. This finding led us to remove contents containing knowledge cutoff dates from the 32B model’s post-training data.",
   "claim": "a false knowledge-cutoff claim traced to post-training examples; the finding changed the 32B model's post-training data",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "uses: fact checking, hallucination, creativity, math",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F33": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'As we start to look towards regulation' (Lebrecht)",
   "quote": "As we start to look towards regulation, governance, and auditability, there are certain key sectors right now that are limited in what they can deploy at scale with black-box models,",
   "claim": "Lebrecht, reported: regulation, governance and auditability limit what key sectors can deploy with black-box models; traceability as a pathway",
   "modality": "attributed",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "uses: regulation and auditing",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F34": {
   "source": "B7",
   "origin": "own_page",
   "locus": "Limitations",
   "quote": "OLMoTrace finds lexical, verbatim matches between an LM’s output and its training data. The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output.",
   "claim": "the authors' stated limit: lexical, verbatim matches only; the documents are not causes of the output, and not supporting evidence or citations for it",
   "modality": "self-description",
   "kind": "function",
   "kernel": true,
   "const": true,
   "lineage": "the stated limit: not causes, not evidence, not citations",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F35": {
   "source": "B3",
   "origin": "card",
   "locus": "PDF, Limitations",
   "quote": "The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output.",
   "claim": "the published version (ACL 2025) carries the same limit",
   "modality": "self-description",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "the stated limit: not causes, not evidence, not citations",
   "date": "2025-07 (ACL 2025 proceedings)",
   "quote_ok": true,
   "checked_against": "saved text B3.txt (sha256 26daccd5e52cd634…), B3-pdf.txt (sha256 f2fc53f52892169d…), whitespace-normalised"
  },
  "F36": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'Ai2 doesn't intend' (Liu)",
   "quote": "Ai2 doesn't intend for the documents retrieved to be viewed as the definitive source of the model's learning. \"We're just showing you some associations, connections between them. Some of them may warrant a further inspection,\" Liu clarified.",
   "claim": "the newsletter and Liu: the documents are not the definitive source of the model's learning, only associations, some warranting inspection",
   "modality": "attributed",
   "kind": "function",
   "kernel": true,
   "const": false,
   "lineage": "the stated limit: not causes, not evidence, not citations",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F37": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'In an example where OLMo was asked about the Space Needle'",
   "quote": "Ai2 asserts that OLMOTrace’s results should not be mistaken as indicators of authority.",
   "claim": "Ai2, reported: the results are not indicators of authority (a matched blog ranks beside an official site)",
   "modality": "attributed",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "the stated limit: not causes, not evidence, not citations",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F38": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'While Ai2 touts OLMoTrace'",
   "quote": "OLMoTrace doesn't provide analytics or insights to help developers determine what to do next to improve the training data. Simply put, it's generating awareness.",
   "claim": "the newsletter's own assessment: it tags where data came from and gives no analytics for improving the training data; it generates awareness",
   "modality": "interpretation",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "assessment from outside",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F39": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§6 'Comparison with RAG.'",
   "quote": "Examples of them include Bing Chat, Google AI Overview, and Perplexity AI. Despite looking similar, OLMoTrace is fundamentally different from RAG: OLMoTrace retrieves documents post-hoc and does not intervene with the LM generation.",
   "claim": "set against retrieval-augmented generation, Google AI Overview named among its examples: OLMoTrace retrieves post hoc and does not intervene in generation",
   "modality": "stipulation",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "unlike retrieval-augmented generation",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F40": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'Liu interjects' (Liu)",
   "quote": "We are not intervening with the model response. Our response is purely generated by the neural language model.",
   "claim": "Liu, reported: unlike Perplexity's retrieval, OLMoTrace does not intervene in the model's response",
   "modality": "attributed",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "unlike retrieval-augmented generation",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F41": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§6 'Other types of tracing.'",
   "quote": "the Gemini App has a “double-check response” feature that highlights parts of the LM response and shows similar results from Google Search, which is updated in real time and thus not identical to Gemini’s training data, making it less useful for scientific exploration.",
   "claim": "Gemini's 'double-check response' shows Google Search results that are not Gemini's training data, less useful for scientific exploration",
   "modality": "interpretation",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "unlike retrieval-augmented generation",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F42": {
   "source": "B7",
   "origin": "own_page",
   "locus": "affiliations",
   "quote": "α Allen Institute for AI ω University of Washington β UC Berkeley σ Stanford University",
   "claim": "the authors' institutions: the Allen Institute for AI, the University of Washington, UC Berkeley, Stanford",
   "modality": "documented",
   "kind": "attribution",
   "kernel": false,
   "const": false,
   "lineage": "the operator and the authors",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F43": {
   "source": "B2",
   "origin": "card",
   "locus": "title and authors",
   "quote": "View a PDF of the paper titled OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens, by Jiacheng Liu and 30 other authors",
   "claim": "the paper: Jiacheng Liu and 30 other authors",
   "modality": "documented",
   "kind": "attribution",
   "kernel": false,
   "const": false,
   "lineage": "the operator and the authors",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F44": {
   "source": "B1",
   "origin": "card",
   "locus": "¶1",
   "quote": "OLMoTrace is a manifestation of Ai2’s commitment to an open ecosystem – open models, open data, and beyond.",
   "claim": "Ai2 presents it as a manifestation of its commitment to open models and open data",
   "modality": "self-description",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "openness",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B1.txt (sha256 7d7b29ab8ba6c6e1…), whitespace-normalised"
  },
  "F45": {
   "source": "B2",
   "origin": "card",
   "locus": "Abstract s6",
   "quote": "OLMoTrace is publicly available and fully open-source.",
   "claim": "publicly available and fully open-source",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "openness",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F46": {
   "source": "B7",
   "origin": "own_page",
   "locus": "§1 ¶5",
   "quote": "The core part of the system is open-sourced under the Apache 2.0 license.",
   "claim": "the core part of the system open-sourced under Apache 2.0",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "openness",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F47": {
   "source": "B6",
   "origin": "card",
   "locus": "¶ 'OLMoTrace connects the output to the input' (Farhadi)",
   "quote": "OLMoTrace connects the output to the input, as simple as that. Because our inputs are open, we could actually connect them,",
   "claim": "Farhadi, reported: it connects output to input, possible because Ai2's inputs are open",
   "modality": "attributed",
   "kind": "interpretation",
   "kernel": false,
   "const": false,
   "lineage": "openness",
   "date": "2025-04-09 (post date)",
   "quote_ok": true,
   "checked_against": "saved text B6.txt (sha256 79b8019f592e2b12…), whitespace-normalised"
  },
  "F48": {
   "source": "B2",
   "origin": "card",
   "locus": "submission history; Comments",
   "quote": "[Submitted on 9 Apr 2025 (v1), last revised 8 Jul 2025 (this version, v2)] … Comments: ACL 2025 demo track",
   "claim": "submitted 9 April 2025, revised 8 July 2025; ACL 2025 demo track",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "publication",
   "date": "2025-04-09 (v1; this page v2, 2025-07-08)",
   "quote_ok": true,
   "checked_against": "saved text B2.txt (sha256 836bf63640ea080a…), whitespace-normalised"
  },
  "F49": {
   "source": "B3",
   "origin": "card",
   "locus": "Anthology metadata",
   "quote": "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) … Award: Best Demo",
   "claim": "published in the ACL 2025 System Demonstrations volume (2025.acl-demo.18), with the Best Demo award",
   "modality": "documented",
   "kind": "fact",
   "kernel": false,
   "const": false,
   "lineage": "publication",
   "date": "2025-07 (ACL 2025 proceedings)",
   "quote_ok": true,
   "checked_against": "saved text B3.txt (sha256 26daccd5e52cd634…), B3-pdf.txt (sha256 f2fc53f52892169d…), whitespace-normalised"
  },
  "F50": {
   "source": "B7",
   "origin": "own_page",
   "locus": "Mitigating social and legal risks",
   "quote": "OLMoTrace can make potentially problematic contents in the LM training data more easily exposed.",
   "claim": "a stated risk: problematic content in the training data more easily exposed; red-teamed for copyright, PII and toxicity",
   "modality": "self-description",
   "kind": "function",
   "kernel": false,
   "const": false,
   "lineage": "risks",
   "date": "2025-07-08 (v2)",
   "quote_ok": true,
   "checked_against": "saved text B7.txt (sha256 0d848290949c7552…), whitespace-normalised"
  },
  "F51": {
   "source": "B8",
   "origin": "wikipedia",
   "locus": "Wikipedia: titles OLMoTrace and OlmoTrace; article-space search",
   "quote": "raw wikitext (action=raw, redirect=no), 2026-10-29T17:09:54Z: HTTP 404 … Wikipedia does not have an article with this exact name. … There were no results matching the query.",
   "claim": "Wikipedia has no article on OLMoTrace under either spelling, no redirect, and no article text containing the name",
   "modality": "documented",
   "kind": "fact",
   "kernel": true,
   "const": false,
   "lineage": "reception: the encyclopedia",
   "date": "2026-10-29 (as fetched)",
   "quote_ok": true,
   "checked_against": "saved text B8.txt (sha256 e0353e047d8f2fa2…), whitespace-normalised"
  }
 },
 "field_sources": {
  "B1": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
  "B2": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
  "B3": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
  "B4": "X — Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
  "B5": "Techzine Global — Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
  "B6": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
  "B7": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
  "B8": "Wikipedia — OLMoTrace (no article)"
 },
 "delta": {
  "T_vs_LB": "At 'OLMoTrace' (T1, six cards: B1–B6, B4 and B5 unreachable; 207 words) the composition gives the tool by its authors' abstract and launch post. Composed from its cards: the definition at the abstract's and the post's grain (B2 F1, B1 F2: tracing outputs back to multi-trillion-token training data in real time; 'full' dropped), Ai2 as maker (B1) with the partners left out (B2 F43 names 31 authors; the affiliations are on B7 F42, the paper the cards link to: CAP-1-05); open source (B2 F45); the button and the verbatim spans (B5 F12, snippet; B1 F4); the index and its complexity (B1 F9); up to 10 documents ranked by BM25 (B1, the post's text of the pipeline; B7 F10); the models (B1 F17, B4 F18 for OLMo 3, the X snippet, CAP-1-06); the uses at the abstract's and the newsletter's grain (B2 F21 'fact checking, hallucination'; B6 F28 'offering fact-checking'; B1 F31 memorization; B6 F33 regulation and auditability). Force raised: the fact-checking use. The field states it as a possibility, 'users may be able to fact-check the statement against its training data' (B3 F24, the card's own paper; B7 F23), qualified by context and source (B7 F25), conditioned in the operator's post on the user personally verifying that the documents 'come from trusted sources' (B1 F27) and called 'quite nuanced' (B1 F26); T composes 'Verify whether specific claims in a model's answer are backed by real training documents', an imperative of support. Limit dropped: the authors' limit, in the card's own paper, that the documents are not 'supporting evidence or citations for the LM output' (B3 F35; B7 F34), and Ai2's statements to the newsletter that they are associations and not 'indicators of authority' (B6 F36, F37): none is composed. The strongest form of the use in the field is Liu's launch statement, users 'can ensure that the data supporting their outputs is trustworthy and verifiable' (B6 F29, attributed); T's 'backed by' stands nearer to it than to the paper, at no attribution. The capture's findings name the same move at their own grain: 'A verbatim match is offered as support for a claim, the reading the paper excludes' (CAP-1-04), 'the conversion of a matching relation into a citation relation that the paper names and refuses' (CAP-1-09). Read against the field whole, fact checking is the authors' own first use (B2 F21; B7 F22) and what the paper excludes is reading the matched documents as causes, evidence or citations: the excess of T is the support relation and the dropped limit, not the use. Substituted reading: T's 'ranks up to 10 context documents using BM25 and unigram probabilities' joins two steps the field keeps apart (the span unigram probability filters spans, B1 'OLMoTrace behind the scenes' ¶4; BM25 ranks documents, B7 F10). Outside the cards: 'such as the Dolma datasets' (no field source names Dolma; the paper's own page, not surfaced, names olmo-mix-1124 and dolmino-mix-1124); 'or generated from general parameters' (no field source makes the dichotomy; the archive's typology of parametric claims, O95-01, bears on it); 'mixes up facts' and 'data recipes and pretraining paths' (no field source). Available, not composed: the count (B1 F16), the purpose (B1 F7), the early stage (B1 F8), the hedge on relevance (B1 F11), the condition of access and the closed providers (B1 F17 in part; B6 F20, F47), the knowledge-cutoff finding (B1 F32), Farhadi's conditional (B6 F30), the contrast with retrieval (B6 F40), the openness as commitment (B1 F44), the publication and its Best Demo award (B2 F48, B3 F49), the newsletter's assessment (B6 F38). Composed without attribution: everything; no person is named, at the coarse grain of a tool description (recorded, not a breach), except the use, which is composed at the grain of a support relation no card states for the documents (recorded as force raised). Content replaced by use: the closer (T1-9).",
  "T_vs_LB_reading": "v0.9 §0.5: the first diff is ontological and narrow. T and the field agree on the entity (Ai2's tool that finds training text verbatim in a model's answer). T composes the tool and its uses and leaves out the one sentence its authors attach to what the tool returns: that a match is not a cause, not evidence, not a citation. With the limit gone, the matching relation is composed as a support relation.",
  "LB_vs_LBA": "Admission adds, in one claim space with the field's: the class's relation, training/output and not output/reader, where PER operates (O725-01); partial tracing possible at cost, recovering no authorship, what survives depending on what is recorded and computed (O1714-01), provenance that has to survive the training pipeline (O1714-02); training-data attribution as the post-hoc detection layer, developed in isolation from the other survival layers (O127-01); the archive's evidentiary hierarchy for training-data attribution, with unvalidated similarity as exploratory evidence only (O1075-01, O1075-02), set beside the paper's heuristic of lexical overlap (F5); parametric claims, which carry no surface signal of their sources and need training-data provenance records at the substrate level (O95-01); the provenance census record's answer to its organizing question, field by field with its basis (PE-01–PE-21), its tag 'traces only models whose training data is open' held against the field's condition of access (PE-17 against F17, F19); and the Capture Registry's account of the composition at the name (CAP-1-01–CAP-1-11). Not surfaced: every archive source; no card cites the census record (CAP-1-11).",
  "alignment": {
   "T1_cards_claims": [
    "F1",
    "F2",
    "F3",
    "F4",
    "F7",
    "F8",
    "F9",
    "F11",
    "F12",
    "F16",
    "F17",
    "F18",
    "F20",
    "F21",
    "F24",
    "F26",
    "F27",
    "F28",
    "F29",
    "F30",
    "F31",
    "F32",
    "F33",
    "F35",
    "F36",
    "F37",
    "F38",
    "F40",
    "F43",
    "F44",
    "F45",
    "F47",
    "F48",
    "F49"
   ],
   "T1_composed": [
    "F1",
    "F2",
    "F3",
    "F4",
    "F9",
    "F10",
    "F12",
    "F17",
    "F18",
    "F21",
    "F24",
    "F28",
    "F31",
    "F33",
    "F45"
   ],
   "T1_outside_cards": [
    "'such as the Dolma datasets' (no field source; B7, own page not surfaced, names olmo-mix-1124 and dolmino-mix-1124)",
    "'or generated from general parameters' (no field source)",
    "'mixes up facts' (no field source)",
    "'data recipes and pretraining paths' (no field source)"
   ],
   "rule": "a field claim counts as composed when T states it at its source's grain or coarser; the typology of what is not is in T_vs_LB (§6.2)"
  }
 },
 "kernel": [
  {
   "K": "K1",
   "claim": "`O725-01`",
   "source": "#725",
   "M_src": "interpretation",
   "sense": "two relations: training/output and output/reader",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K2",
   "claim": "`O95-01`",
   "source": "#95",
   "M_src": "interpretation",
   "sense": "parametric claims and training-data provenance",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K3",
   "claim": "`O1075-01`",
   "source": "#1075",
   "M_src": "interpretation",
   "sense": "the evidentiary hierarchy for training-data attribution",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F5 (the paper's heuristic of lexical overlap as likely influence)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K4",
   "claim": "`O1075-02`",
   "source": "#1075",
   "M_src": "definition",
   "sense": "the evidentiary hierarchy for training-data attribution",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F5 (the paper's heuristic of lexical overlap as likely influence)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K5",
   "claim": "`O127-01`",
   "source": "#127",
   "M_src": "interpretation",
   "sense": "post-hoc detection: one survival layer, developed in isolation",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K6",
   "claim": "`O1714-01`",
   "source": "#1714",
   "M_src": "interpretation",
   "sense": "partial tracing at cost; what survives depends on what is recorded",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F5 (the paper's heuristic of lexical overlap as likely influence)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K7",
   "claim": "`O1714-02`",
   "source": "#1714",
   "M_src": "interpretation",
   "sense": "partial tracing at cost; what survives depends on what is recorded",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K8",
   "claim": "`CAP-1-01`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "the composition at the name: what it held",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K9",
   "claim": "`CAP-1-02`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "the composition at the name: what it held",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K10",
   "claim": "`CAP-1-03`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "the stated limit: not causes, not evidence, not citations",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K11",
   "claim": "`CAP-1-04`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "interpretation",
   "sense": "the composition converts a match into support",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F21, F22, F23, F28, F29 (fact checking as the field states it)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K12",
   "claim": "`CAP-1-05`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "the operator and the authors",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K13",
   "claim": "`CAP-1-06`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "models and applicability",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K14",
   "claim": "`CAP-1-07`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "stipulation",
   "sense": "models and applicability",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K15",
   "claim": "`CAP-1-08`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "interpretation",
   "sense": "the composition converts a match into support",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F21, F22, F23, F28, F29 (fact checking as the field states it)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K16",
   "claim": "`CAP-1-09`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "interpretation",
   "sense": "the composition converts a match into support",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "qualified claim: F21, F22, F23, F28, F29 (fact checking as the field states it)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K17",
   "claim": "`CAP-1-10`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "interpretation",
   "sense": "the composition at the name: what it held",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K18",
   "claim": "`CAP-1-11`",
   "source": "Capture Registry, 2026-10-29",
   "M_src": "documented",
   "sense": "reception: the record and the cards",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing finding (an observation of the composition at the entity's address)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K19",
   "claim": "`PE-01`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "what it is: tracing model output to its training data in real time",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K20",
   "claim": "`PE-02`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "what it is: tracing model output to its training data in real time",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K21",
   "claim": "`PE-03`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the relation: verbatim match between output and training text",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K22",
   "claim": "`PE-04`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the relation: verbatim match between output and training text",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K23",
   "claim": "`PE-05`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the pipeline: spans, index, retrieval, ranking",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K24",
   "claim": "`PE-06`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the pipeline: spans, index, retrieval, ranking",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K25",
   "claim": "`PE-07`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "attributed",
   "sense": "the training data and its count",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K26",
   "claim": "`PE-08`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "attributed",
   "sense": "the operator and the authors",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K27",
   "claim": "`PE-09`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "what the record's schema asks",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K28",
   "claim": "`PE-10`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the relation: verbatim match between output and training text",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K29",
   "claim": "`PE-11`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "attributed",
   "sense": "the stated limit: not causes, not evidence, not citations",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "none (restates F34/F35, one lineage)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K30",
   "claim": "`PE-12`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "attributed",
   "sense": "openness",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K31",
   "claim": "`PE-13`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "documented",
   "sense": "publication",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K32",
   "claim": "`PE-14`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "documented",
   "sense": "publication",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K33",
   "claim": "`PE-15`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "attributed",
   "sense": "the stated limit: not causes, not evidence, not citations",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "none (restates F34/F35, one lineage)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K34",
   "claim": "`PE-16`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "the stated limit: not causes, not evidence, not citations",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "none (restates F34/F35, one lineage)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K35",
   "claim": "`PE-17`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "models and applicability",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "rival claim: F17, F19 (the condition is access to the full training data)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K36",
   "claim": "`PE-18`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "neighbours in the census",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K37",
   "claim": "`PE-19`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "self-description",
   "sense": "what the record's schema asks",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K38",
   "claim": "`PE-20`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "self-description",
   "sense": "what the record's schema asks",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  },
  {
   "K": "K39",
   "claim": "`PE-21`",
   "source": "provenance census record, 2026-10-29",
   "M_src": "stipulation",
   "sense": "what the record's schema asks",
   "qualifiers carried": "—",
   "f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the claim, attributed or not, uptake §8.3)",
   "contrast": "missing distinction (the field has no claim on the point)",
   "resolver": "a later capture at 'OLMoTrace' (ADDR-435813c322c8) on Google AI Overview, seated in the Capture Registry (§8.3)"
  }
 ],
 "kernel_T": [
  {
   "claim": "F35",
   "note": "the authors' limit on the card's own paper (not causes, not supporting evidence or citations), absent"
  },
  {
   "claim": "F24",
   "note": "fact checking as a possibility ('may be able to'), composed as an imperative of support"
  },
  {
   "claim": "F26",
   "note": "'quite nuanced', absent"
  },
  {
   "claim": "F27",
   "note": "confidence conditional on the user's verification of the sources, absent"
  },
  {
   "claim": "F36",
   "note": "associations, not the definitive source, absent"
  },
  {
   "claim": "F37",
   "note": "not indicators of authority, absent"
  },
  {
   "claim": "F29",
   "note": "Liu's launch statement, the strongest form of the use, unattributed in T's grain"
  },
  {
   "claim": "F11",
   "note": "the hedge on BM25 ('some notion of relevance'), absent"
  },
  {
   "claim": "F16",
   "note": "the count, absent"
  },
  {
   "claim": "F17",
   "note": "the condition of access to the training data, absent"
  },
  {
   "claim": "F20",
   "note": "closed-model providers' hesitation, absent"
  },
  {
   "claim": "F47",
   "note": "openness as the condition, absent"
  },
  {
   "claim": "F7",
   "note": "the purpose, absent"
  },
  {
   "claim": "F8",
   "note": "the early stage, absent"
  },
  {
   "claim": "F32",
   "note": "the knowledge-cutoff finding, absent"
  },
  {
   "claim": "F30",
   "note": "Farhadi's conditional on factual inputs, absent"
  },
  {
   "claim": "F40",
   "note": "the contrast with retrieval, absent"
  },
  {
   "claim": "F44",
   "note": "openness as commitment, absent"
  },
  {
   "claim": "F43",
   "note": "the authors, absent (Ai2 alone)"
  },
  {
   "claim": "F48",
   "note": "publication dates, absent"
  },
  {
   "claim": "F49",
   "note": "ACL 2025 and the Best Demo award, absent"
  },
  {
   "claim": "F38",
   "note": "the newsletter's assessment, absent"
  }
 ],
 "ledger": {
  "archive": "datasets/negative-of-the-negative/v2/ledgers/olmotrace/ledger-archive.json",
  "field": "datasets/negative-of-the-negative/v2/ledgers/olmotrace/field.json",
  "selection": "datasets/negative-of-the-negative/v2/traversal/olmotrace/",
  "config": "datasets/negative-of-the-negative/v2/panel/configs/olmotrace.json",
  "intake": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/",
  "audit": null
 },
 "ontology": {
  "composition": "equal",
  "composition_note": "equal-terms case (v0.9 §0.9, §4.1): one claim space; the field's and the archive's claims ordered together by the entity's dependency and by genealogy within each sense; nothing translated; PE-17 against F17/F19 and F29 against F34/F35 held under P_coh",
  "case": {
   "value": "equal",
   "statement": "The archive reads the field's entity as the entity. The provenance census record describes OLMoTrace from its authors' own pages, each field with its basis ('operator'), and answers for it the census's question of what provenance a system carries, through which operation, on whose assertion, with what evidence of survival; the capture record checks the composition against those pages (CAP-1-01–CAP-1-02). No deposit names the tool. The deposits bear on its class, tracing model outputs to training data, and add senses and limits on the field's own object: the training/output relation against PER's output/reader relation (#725), partial tracing at cost that recovers no authorship (#1714), the post-hoc detection layer developed in isolation (#127), similarity as exploratory evidence only (#1075), and the parametric claims such records are companion infrastructure for (#95). #725 states its own relation to the frameworks it places in these terms ('Extension and completion, not substitution or critique'). No defining paper contests the field's ontology of the tool or supplies one the field lacks: ρ = extends, the equal-terms case (§0.9, §1.2a). The archive's reading of the composition, a matching relation converted into a citation relation (CAP-1-09), takes the authors' limit as given and is the field's own distinction carried to the composition layer.",
   "loci": [
    {
     "dep": 725,
     "locus": "packet header, Relation",
     "quote": "**Relation:** Extension and completion, not substitution or critique"
    },
    {
     "dep": 725,
     "locus": "§4 table, row 'Data attribution (TRAK, influence functions)'",
     "quote": "| Data attribution (TRAK, influence functions) | Which training examples shaped a specific output | Adjacent eval method | Operates on training/output relation; PER operates on output/reader relation |"
    },
    {
     "dep": 1714,
     "locus": "§9.2 s4–5",
     "quote": "The attribution literature shows that partial tracing is possible at cost: influence-function methods estimate which training examples most affect a given output (Grosse et al. 2023). Neither recovers authorship, and both show that what survives depends on what is recorded and computed."
    },
    {
     "dep": 95,
     "locus": "the typology of source-dependent claims",
     "quote": "Parametric claims are drawn from the model's training-time parameters rather than runtime retrieval, often without surface signal indicating which sources materially contributed; PER applies to these only indirectly, and addressing them at the substrate level requires companion infrastructure (training-data provenance records, watermarking, content credentials)."
    },
    {
     "dep": "PE",
     "locus": "census.json organizing_question",
     "quote": "What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?"
    },
    {
     "dep": "PE",
     "locus": "census.json rows[slug == 'olmotrace'].provenance_object",
     "quote": "a span of a language model's output and the training documents that contain it verbatim"
    },
    {
     "dep": "capture-registry",
     "locus": "olmotrace-aio-20261029, analysis s2",
     "quote": "The composition turns a trace of verbatim overlap into evidence for a claim, the conversion of a matching relation into a citation relation that the paper names and refuses."
    }
   ]
  },
  "E_F": {
   "type": "C (public entity: a tracing tool and the paper that presents it)",
   "definition": "the system of the Allen Institute for AI with the University of Washington, UC Berkeley and Stanford (Liu, Blanton and 29 others; arXiv:2504.07096, 9 April 2025, v2 8 July 2025; ACL 2025 System Demonstrations, Best Demo) that finds and shows, in the Ai2 Playground and in real time, verbatim matches between long, unique spans of a language model's output and documents in the model's full training data (for OLMo-2-32B-Instruct 3.2 billion documents, 4.6 trillion tokens), over an extended infini-gram index, retrieving up to 10 snippets per span ranked by BM25; its purpose to explore where a model may have learned a word sequence, its uses shown on fact checking (a possibility, conditional on the user's verification of the sources), creativity and math; its stated limit that the documents are not causes of the output, nor supporting evidence or citations for it; set apart from retrieval-augmented generation, Google AI Overview among RAG's examples; applicable to any model whose full training data its provider has access to; open source",
   "claims": [
    "F1",
    "F2",
    "F3",
    "F4",
    "F42",
    "F43",
    "F48",
    "F49",
    "F15",
    "F9",
    "F10",
    "F6",
    "F21",
    "F23",
    "F27",
    "F34",
    "F39",
    "F19",
    "F45"
   ]
  },
  "E_A": {
   "definition": "the same tool, in the provenance census record and the archive's deposits: a tracing tool in the lineage of training data that records provenance, its object a span of model output and the training documents that contain it verbatim, its relation a verbatim match, its carrier the full training corpus, with the authors' limit as its negative tags (not causes; not a citation system); a member of the class of training-data attribution, which operates on the training/output relation where PER operates on the output/reader relation (#725), traces partially and at cost without recovering authorship (#1714), stands as one post-hoc detection layer in isolation (#127), holds similarity as exploratory evidence (#1075), and supplies the training-data provenance records parametric claims need (#95); whose composition at its own name converts the matching relation into a citation relation (Capture Registry, 2026-10-29)",
   "claims": [
    "PE-01",
    "PE-02",
    "PE-09",
    "PE-03",
    "PE-04",
    "PE-05",
    "PE-15",
    "PE-16",
    "O725-01",
    "O1714-01",
    "O127-01",
    "O1075-02",
    "O95-01",
    "CAP-1-09"
   ],
   "defining_papers": [
    725,
    95,
    1075,
    127,
    1714
   ],
   "defining_records": [
    "provenance census record (census.json rows[slug == 'olmotrace'] and organizing_question; rendered aggregators/olmotrace/index.html; editor Lee Sharks, compiled 2026-10-29, schema 1.0)",
    "Capture Registry olmotrace-aio-20261029 (OBS-3fd20ae244c0)"
   ]
  },
  "rho": {
   "value": "extends",
   "status": "read from the provenance census record, the capture record and the deposits that bear on the tool's class (§1.2a); no deposit names the tool; #725 names its relation to the frameworks it places 'Extension and completion, not substitution or critique'",
   "statement": "The archive reads OLMoTrace as its authors present it and adds, on the field's own object: the relation the class operates on (training/output) against the one PER measures (output/reader); the cost and partiality of tracing and what it does not recover; its isolation as one survival layer; the evidentiary grade of similarity; the parametric claims it is companion infrastructure for; the census's answer, field by field, to what provenance it carries; and the composition layer's conversion, at its name, of a match into support.",
   "claims": [
    "O725-01",
    "O1714-01",
    "O1714-02",
    "O127-01",
    "O1075-01",
    "O1075-02",
    "O95-01",
    "PE-21",
    "PE-03",
    "PE-04",
    "CAP-1-04",
    "CAP-1-09"
   ],
   "guard": "ρ = extends claims an addition and no more (§1.2a). It does not make the field's sources wrong at their dates. Where an archive claim stands against a field claim (PE-17 against F17/F19, open or accessible training data), both are composed under P_coh within the one ontology. The archive's reading of fact checking at the composition layer (CAP-1-04, CAP-1-09) is composed beside the field's own statements of the use (F21–F29) and of the limit (F34–F37), not in their place.",
   "considered": [
    {
     "value": "constructs",
     "rejected": "the archive does not read the tool as made by reception; its records take the authors' pages as the entity's own account"
    },
    {
     "value": "revises",
     "rejected": "the tool is not the archive's own entity; the archive has no earlier statement of it to keep, narrow or withdraw"
    }
   ]
  },
  "translation": [
   {
    "lineage": "what it is: tracing model output to its training data in real time",
    "field_claims": [
     "F1",
     "F2"
    ],
    "received": "F1 (B2, Abstract s1): \"We present OLMoTrace, the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time.\"; F2 (B1, ¶1): \"Today we introduce OLMoTrace, a one-of-a-kind feature in the Ai2 Playground that lets you trace the outputs of language models back to their full, multi-trillion-token training data in real time.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-01, PE-02",
    "archive_claims": [
     "PE-01",
     "PE-02"
    ]
   },
   {
    "lineage": "the relation: verbatim match between output and training text",
    "field_claims": [
     "F3",
     "F4",
     "F5"
    ],
    "received": "F3 (B2, Abstract s2): \"OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora.\"; F4 (B1, 'Interact with OLMoTrace' ¶1): \"The highlights indicate long and unique spans that appear verbatim at least once in the training data of this model.\"; F5 (B7, §6 Related Work, 'Tracing LM generation into training data'): \"we directly retrieve similar training examples by lexical overlap, with the heuristic that such training examples are likely to be influential for the given output.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-03, PE-04, PE-10",
    "archive_claims": [
     "PE-03",
     "PE-04",
     "PE-10"
    ]
   },
   {
    "lineage": "purpose: where a model may have learned a word sequence",
    "field_claims": [
     "F6",
     "F7",
     "F8"
    ],
    "received": "F6 (B7, §1 ¶4): \"The purpose of OLMoTrace is to give users a tool to explore where LMs may have learned to generate certain word sequences, focusing on verbatim matching as the most direct connection between LM outputs and the training data.\"; F7 (B1, ¶2): \"We developed OLMoTrace to enable researchers, developers, and the general public to inspect where and how language models may have learned to generate certain word sequences.\"; F8 (B1, 'Interpreting results' ¶5): \"We are still in an early stage in exploring the full potential of this tool\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "the pipeline: spans, index, retrieval, ranking",
    "field_claims": [
     "F9",
     "F10",
     "F11",
     "F12"
    ],
    "received": "F9 (B1, 'OLMoTrace behind the scenes' ¶3): \"reducing the time complexity of span finding from the naive O(L^2 * N) to O(L * log N)\"; F10 (B7, §3 Steps 3 and 5): \"For each kept span, we retrieve up to 10 document snippets from the training data that enclose this span. … we rank all documents by a BM25 score in descending order.\"; F11 (B1, 'OLMoTrace behind the scenes' ¶7): \"We find that this BM25 score conveys some notion of “relevance”, and thus we surface documents with higher scores on top.\"; F12 (B5, card snippet (Techzine unreachable: HTTP 403)): \"After generating an answer from OLMo in the Ai2 Playground, users can activate the tool by clicking the “Show OLMoTrace” button below the output.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-05, PE-06",
    "archive_claims": [
     "PE-05",
     "PE-06"
    ]
   },
   {
    "lineage": "performance and relevance",
    "field_claims": [
     "F13",
     "F14"
    ],
    "received": "F13 (B7, §3.2): \"the OLMoTrace inference latency per query is 4.46 seconds\"; F14 (B7, §4 'Validating the relevance rankings'): \"The first document displayed in each conversation received an average relevance score of 1.90 (roughly meaning “being on the same topic as the LM output)\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "the training data and its count",
    "field_claims": [
     "F15",
     "F16"
    ],
    "received": "F15 (B7, §2 'The training data.'): \"which totals 3.2 billion documents and 4.6 trillion (Llama-2) tokens\"; F16 (B1, 'OLMoTrace behind the scenes' ¶8): \"In total, these five datasets contain about 3.2 billion documents and 4.6 trillion tokens, and OLMoTrace matches against all of them.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-07",
    "archive_claims": [
     "PE-07"
    ]
   },
   {
    "lineage": "models and applicability",
    "field_claims": [
     "F17",
     "F18",
     "F19",
     "F20"
    ],
    "received": "F17 (B1, ¶2): \"OLMoTrace is available today with our three flagship models: OLMo 2 32B Instruct, OLMo 2 13B Instruct, and OLMoE 1B 7B Instruct, and it can be applied to any language model for which you have access to the training data.\"; F18 (B4, card title and snippet (X unreachable)): \"Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just see its behavior—you can trace it. For example, you can ask Olmo 3-Think (32B\"; F19 (B7, §1 ¶5): \"OLMoTrace can be applied to any LM as long as the service provider has access to its full training data.\"; F20 (B6, ¶ 'OLMoTrace isn't limited to Ai2's model'): \"However, Liu reasons that closed model providers may hesitate to use OLMoTrace because it would expose the training data to sunlight.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside CAP-1-06, CAP-1-07, PE-17",
    "archive_claims": [
     "CAP-1-06",
     "CAP-1-07",
     "PE-17"
    ]
   },
   {
    "lineage": "uses: fact checking, hallucination, creativity, math",
    "field_claims": [
     "F21",
     "F22",
     "F30",
     "F31",
     "F32"
    ],
    "received": "F21 (B2, Abstract s5): \"We showcase how it can be used to explore fact checking, hallucination, and the creativity of language models.\"; F22 (B7, §1 ¶4): \"We present three case studies for ways to use OLMoTrace (§ 5 ): (1) fact checking, (2) tracing the LM-generated “creative” expressions, and (3) tracing math capabilities.\"; F30 (B6, ¶ 'OLMoTrace connects the output to the input' (Farhadi)): \"And the minute you start linking the generation to an input—assuming that the inputs were factual—then you can start also cooking up scores for hallucination\"; F31 (B1, 'Interpreting results' ¶4): \"OLMoTrace reveals that this entire expression appears multiple times in the training data, so the model could have easily learned it through memorization.\"; F32 (B1, 'Interpreting results' ¶5): \"now we can see that the model may have learned to say August 2023 from the post-training examples listed on the right. This finding led us to remove contents containing knowledge cutoff dates from the 32B model’s post-training data.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "fact checking: what a matched document can and cannot show",
    "field_claims": [
     "F23",
     "F24",
     "F25",
     "F26",
     "F27",
     "F28",
     "F29"
    ],
    "received": "F23 (B7, §5 'Fact checking.'): \"If the LM states a fact, users may be able to fact-check the statement against its training data.\"; F24 (B3, PDF §5 'Fact checking.'): \"If the LM states a fact, users may be able to fact-check the statement against its training data.\"; F25 (B7, §5 'Fact checking.' ¶2): \"We note that inspecting the document context and source can help users make a more informed judgment on the factuality of the statement, as words can be misleading out of context, and some web sources may be unreliable.\"; F26 (B1, 'Interpreting results' ¶1): \"The interpretation of text matches with training data is quite nuanced. For some highlighted spans in the model response, we can use their corresponding documents to fact check claims made in the spans.\"; F27 (B1, 'Interpreting results' ¶2): \"If you personally verify that these documents come from trusted sources, you can gain confidence that this piece of information generated by the model is factually correct.\"; F28 (B6, ¶1): \"Called OLMoTrace, it's an open-source application offering fact-checking of information provided in prompt replies.\"; F29 (B6, ¶2 (Liu, in a statement)): \"anyone using our models can ensure that the data supporting their outputs is trustworthy and verifiable.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "uses: regulation and auditing",
    "field_claims": [
     "F33"
    ],
    "received": "F33 (B6, ¶ 'As we start to look towards regulation' (Lebrecht)): \"As we start to look towards regulation, governance, and auditability, there are certain key sectors right now that are limited in what they can deploy at scale with black-box models,\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "the stated limit: not causes, not evidence, not citations",
    "field_claims": [
     "F34",
     "F35",
     "F36",
     "F37"
    ],
    "received": "F34 (B7, Limitations): \"OLMoTrace finds lexical, verbatim matches between an LM’s output and its training data. The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output.\"; F35 (B3, PDF, Limitations): \"The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output.\"; F36 (B6, ¶ 'Ai2 doesn't intend' (Liu)): \"Ai2 doesn't intend for the documents retrieved to be viewed as the definitive source of the model's learning. \"We're just showing you some associations, connections between them. Some of them may warrant a further inspection,\" Liu clarified.\"; F37 (B6, ¶ 'In an example where OLMo was asked about the Space Needle'): \"Ai2 asserts that OLMOTrace’s results should not be mistaken as indicators of authority.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside CAP-1-03, PE-11, PE-15, PE-16",
    "archive_claims": [
     "CAP-1-03",
     "PE-11",
     "PE-15",
     "PE-16"
    ]
   },
   {
    "lineage": "assessment from outside",
    "field_claims": [
     "F38"
    ],
    "received": "F38 (B6, ¶ 'While Ai2 touts OLMoTrace'): \"OLMoTrace doesn't provide analytics or insights to help developers determine what to do next to improve the training data. Simply put, it's generating awareness.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "unlike retrieval-augmented generation",
    "field_claims": [
     "F39",
     "F40",
     "F41"
    ],
    "received": "F39 (B7, §6 'Comparison with RAG.'): \"Examples of them include Bing Chat, Google AI Overview, and Perplexity AI. Despite looking similar, OLMoTrace is fundamentally different from RAG: OLMoTrace retrieves documents post-hoc and does not intervene with the LM generation.\"; F40 (B6, ¶ 'Liu interjects' (Liu)): \"We are not intervening with the model response. Our response is purely generated by the neural language model.\"; F41 (B7, §6 'Other types of tracing.'): \"the Gemini App has a “double-check response” feature that highlights parts of the LM response and shows similar results from Google Search, which is updated in real time and thus not identical to Gemini’s training data, making it less useful for scientific exploration.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "the operator and the authors",
    "field_claims": [
     "F42",
     "F43"
    ],
    "received": "F42 (B7, affiliations): \"α Allen Institute for AI ω University of Washington β UC Berkeley σ Stanford University\"; F43 (B2, title and authors): \"View a PDF of the paper titled OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens, by Jiacheng Liu and 30 other authors\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside CAP-1-05, PE-08",
    "archive_claims": [
     "CAP-1-05",
     "PE-08"
    ]
   },
   {
    "lineage": "openness",
    "field_claims": [
     "F44",
     "F45",
     "F46",
     "F47"
    ],
    "received": "F44 (B1, ¶1): \"OLMoTrace is a manifestation of Ai2’s commitment to an open ecosystem – open models, open data, and beyond.\"; F45 (B2, Abstract s6): \"OLMoTrace is publicly available and fully open-source.\"; F46 (B7, §1 ¶5): \"The core part of the system is open-sourced under the Apache 2.0 license.\"; F47 (B6, ¶ 'OLMoTrace connects the output to the input' (Farhadi)): \"OLMoTrace connects the output to the input, as simple as that. Because our inputs are open, we could actually connect them,\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-12",
    "archive_claims": [
     "PE-12"
    ]
   },
   {
    "lineage": "publication",
    "field_claims": [
     "F48",
     "F49"
    ],
    "received": "F48 (B2, submission history; Comments): \"[Submitted on 9 Apr 2025 (v1), last revised 8 Jul 2025 (this version, v2)] … Comments: ACL 2025 demo track\"; F49 (B3, Anthology metadata): \"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) … Award: Best Demo\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received beside PE-13, PE-14",
    "archive_claims": [
     "PE-13",
     "PE-14"
    ]
   },
   {
    "lineage": "risks",
    "field_claims": [
     "F50"
    ],
    "received": "F50 (B7, Mitigating social and legal risks): \"OLMoTrace can make potentially problematic contents in the LM training data more easily exposed.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   },
   {
    "lineage": "reception: the encyclopedia",
    "field_claims": [
     "F51"
    ],
    "received": "F51 (B8, Wikipedia: titles OLMoTrace and OlmoTrace; article-space search): \"raw wikitext (action=raw, redirect=no), 2026-10-29T17:09:54Z: HTTP 404 … Wikipedia does not have an article with this exact name. … There were no results matching the query.\"",
    "receiving_text": "—",
    "translation": "untranslated: composed as received (no archive claim in its lineage)",
    "archive_claims": []
   }
  ],
  "receiving_texts": "equal-terms case: every field claim composed as received, beside the archive claims it stands with"
 },
 "measure": {
  "rule": "v0.9 §4.10: KO_B sentences surviving in the KO at similarity ≥ 0.95, with positions (computed as apply_v09.py does: difflib.SequenceMatcher ratio, the best KO sentence for each KO_B sentence)",
  "v0.9": {
   "surviving": 0,
   "of": 14,
   "positions": [
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null,
    null
   ]
  },
  "opening_run": {
   "rule": "v0.9 §5.10 (equal-terms case): the field-only KO's sentences as the KO's opening run in their own order",
   "breach": false
  },
  "blocks": {
   "rule": "v0.9 §4.1 (equal-terms and mixed cases): the KO does not open with the field as a block or close with the archive as one; sentences typed by the ledgers of the claims they cite (F field-only, A archive-only, M both)",
   "sequence": "MMMMMFFMMFAAAFFFMFFFAAMMMFM",
   "field_only_head": 0,
   "archive_only_tail": 0,
   "breach": false
  },
  "max_similarity": [
   0.498,
   0.196,
   0.604,
   0.618,
   0.648,
   0.491,
   0.493,
   0.166,
   0.325,
   0.541,
   0.623,
   0.486,
   0.537,
   0.263
  ],
  "ko_sentence_1_max_similarity_to_KO_B": 0.498
 },
 "homonyms": {
  "names": [],
  "basis": "The census row carries no disambiguation field. No card is a homonym. 'OlmoTrace' (the X card) and 'OLMOTRACE' (the ACL abstract's small capitals) are spellings of the entity. The distinct entities below are named by the field or the census beside the tool and are never senses of its name.",
  "distinct_entities": [
   "infini-gram (the index the tool extends; the census lists it as unchecked, 'infini-gram as a separate record')",
   "the Ai2 Playground (the surface the tool runs on; unchecked in the census)",
   "the OLMo models (OLMo 2, OLMoE, OLMo 3; the models it traces)",
   "Dolma (named by T, by no field source; the field names olmo-mix-1124 and dolmino-mix-1124)",
   "influence functions, TRAK and training-data attribution methods (the class's other methods: B7 §6; #725, #1075, #1714)",
   "retrieval-augmented generation systems, Google AI Overview, Bing Chat and Perplexity among them, and Gemini's 'double-check response' (set apart by B7 §6 and B6)",
   "the Data Provenance Initiative (the census's neighbour; its /non row is datasets/negative-of-the-negative/v2/rows/data-provenance-explorer.json, not edited)"
  ],
  "substitutes": {
   "T": []
  },
  "register": {
   "address_1": {
    "slug": "olmotrace-aio-20261029",
    "obs_id": "OBS-3fd20ae244c0"
   }
  }
 },
 "notes": [
  "One address, Google AI Overview (expanded), signed out, incognito, 2026-10-29: olmotrace-aio-20261029 (OBS-3fd20ae244c0, ADDR-435813c322c8, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔), six cards, 207 words in the answer body with its closer. A NEW address; the census record landed the same day and no card cites it (CAP-1-11).",
  "The limit, checked against the fetched cards as the task asks. The authors' sentence, 'The retrieved documents should not be interpreted as having a causal effect on the LM output, or as supporting evidence or citations for the LM output', stands in the paper's Limitations in both fetched versions: arXiv HTML v2 (B7, the census's own page, F34) and the ACL 2025 Anthology PDF behind card 3 (B3, F35). It is not on the abstract page (card 2, B2), nor in Ai2's blog post (card 1, B1), which instead calls interpretation 'quite nuanced' and conditions confidence on the user personally verifying that the documents 'come from trusted sources' (F26, F27). The newsletter (card 6, B6) carries it in attenuated, attributed form ('associations', not 'the definitive source', not 'indicators of authority': F36, F37), and also carries its opposite in Ai2's launch statement (Liu: users 'can ensure that the data supporting their outputs is trustworthy and verifiable', F29). The field holds fact checking as the authors' own first use (abstract, F21; the paper's first case study, F22), stated as a possibility ('users may be able to fact-check the statement against its training data', F23, F24) with a qualification (F25). So the field holds the use and the limit together, within the operator's own family, and the row composes both under P_coh. T's 'Verify whether specific claims … are backed by real training documents' raises the force of the use and drops the limit; it is not a use the authors disclaim.",
  "Fetching: every reachable page saved verbatim with curl on 2026-10-29 between 17:07 and 17:10 UTC, raw and text with sha256 in field.json. Card URLs: B1, B3 and B5 by WebSearch on title and site; B2 from the snippet's arXiv number (byte-identical to the census's saved copy olmotrace--abs, sha256 cd982680…, so carried once, as a card); B6 from the newsletter's archive API. B7 copied from the census's saved copy olmotrace--paper (sha256 matches the census record); not refetched. The ACL PDF (B3-pdf.pdf) was fetched beside the Anthology page to read the card's own paper for its limit.",
  "Unreachable: B4, Ai2's post on X: no status URL could be found (three searches on the title), and x.com serves an unauthenticated fetch only a script shell (the profile and a search page tried, both saved, neither carrying the post); B5, Techzine: HTTP 403 on three fetches with the brief's command (a fetch with other request headers was refused by the workspace as bot-check circumvention and not pursued). Their titles and snippets stand (F12, F18).",
  "Card 1's snippet ('Feature: OLMoTrace in Ai2 Playground traces language model outputs to multi-trillion-token training data in real time. Supported Models: OLMo 2 32B Instruct, OL…') is not on the page the card's title names: neither 'Feature:' nor 'Supported Models' occurs in the served blog post, whose meta description is Ai2's generic one. The snippet's text is not the page's; recorded, not resolved, and not ledgered as a field claim.",
  "The archive subset: no archive text names OLMoTrace, its arXiv number, its Anthology identifier or infini-gram (D 0, R 0). The string pass returned 4 O-class candidates (#73, #580, #623, #1334), all rejected by reading. Because the O-class pass returns a sentence only where it also carries a term of the script's category vocabulary, every archive sentence matching either class pattern (14 in 12 texts) was read; that reading admitted #725, #95, #1075, #127 and #1714, with #1190 and #1188 grouped as instances (S4). The hop (77 candidates in 76 deposits) admitted none. Defect of the string pass, reported, not fixed: the five admitted deposits were all missed by the O-class pass, as #95 was on the Data Provenance Explorer row.",
  "Defects in the provenance census record (reported, not fixed): (a) negative_tags[2] 'traces only models whose training data is open' is contradicted by the field: the paper states the condition as access, 'OLMoTrace can be applied to any LM as long as the service provider has access to its full training data' (B7 §1, F19), and the operator's post 'any language model for which you have access to the training data' (B1, F17); the newsletter reports it can be applied to OpenAI, Anthropic and Meta models, closed providers hesitating (B6, F20). (b) The count is OLMo-2-32B-Instruct's (paper v2); the capture shows the composition carrying the OLMo 3 family from Ai2's post on X (CAP-1-06, CAP-1-07): the record is out of date on models, and no Ai2 page listing OLMo 3 under the tool was found in this row's field (the X post was unreachable). (c) data_access 'open source' and the evidence quote 'publicly available and fully open-source' (abstract) stand beside the paper body's 'The core part of the system is open-sourced under the Apache 2.0 license' (F46): not a contradiction of the field, but the record carries only the stronger form. (d) lifecycle.detail 'ACL 2025 demo track' omits what the Anthology page states: published in the System Demonstrations volume, pages 178–188, with the Best Demo award (B3, F49). (e) unit 'a maximal matching span of model output' is the pipeline's step 1; the paper's displayed unit is the span kept by unigram-probability filtering and merged with overlapping spans (B7 §3 steps 2 and 4; B1). (f) The census row has no disambiguation field (as on the MMRS rows).",
  "Observations on the capture (reported, not fixed): findings[1] ('THE DISCLAIMED USE GIVEN AS A USE') and the capture's title ('THE USE ITS AUTHORS DISCLAIM') say more than the field supports: fact checking is the authors' own first use and case study (F21, F22, F23), and Ai2's post gives it with its condition (F27); what the paper disclaims is reading the matched documents as causes, supporting evidence or citations (F34, F35). The finding's second half ('A verbatim match is offered as support for a claim, the reading the paper excludes') and the analysis ('the conversion of a matching relation into a citation relation') are borne out. The partners and the OLMo 3 findings are borne out by the field (F42, F43; F18).",
  "Related rows: the census names the Data Provenance Initiative as the entity's neighbour; its /non row is datasets/negative-of-the-negative/v2/rows/data-provenance-explorer.json (not edited). The paper names Google AI Overview as a RAG system from which it sets OLMoTrace apart (F39); the AIO row is datasets/negative-of-the-negative/v2/rows/google-ai-overviews.json (not edited)."
 ],
 "composition_v09": {
  "source": "this row (composed directly; no rebuild/negative-of-the-negative/v09/olmotrace.json)",
  "composer": "Claude (Opus 5.5), subagent, 2026-10-29",
  "defining_material_read": [
   {
    "what": "provenance census record, row",
    "file": "/home/claude/fleet/provenance-erasure/data/provenance-census/census.json (rows[slug == 'olmotrace'], organizing_question, schema_note)",
    "read": "whole"
   },
   {
    "what": "provenance census record, rendered",
    "file": "/home/claude/fleet/provenance-erasure/aggregators/olmotrace/index.html",
    "read": "text, for agreement with the row"
   },
   {
    "what": "census sources",
    "file": "/home/claude/fleet/provenance-erasure/data/provenance-census/sources/olmotrace--abs.*, olmotrace--paper.*",
    "read": "whole (the paper's text read through)"
   },
   {
    "what": "capture record",
    "file": "data/EA-WG-CAPTURES-01.json (olmotrace-aio-20261029)",
    "read": "transcript, transcript_raw, cite_list, d, reading, findings, analysis, q_kind, per_note, originator"
   },
   {
    "dep": 725,
    "file": "data/texts/AXN-0273-text.md",
    "read": "the header, §4's tables (provenance frameworks and adjacent measurement concepts) and §5.1"
   },
   {
    "dep": 95,
    "file": "data/texts/AXN-026A-text.md",
    "read": "the PER levels and the typology of source-dependent claims"
   },
   {
    "dep": 1075,
    "file": "data/texts/AXN-0444-text.md",
    "read": "the paragraph on training-data attribution research and §9.1–9.2"
   },
   {
    "dep": 127,
    "file": "data/texts/AXN-02B2-text.md",
    "read": "Gap 2"
   },
   {
    "dep": 1714,
    "file": "data/texts/AXN-0A7F-text.md",
    "read": "§9.1–9.3 and the revision note naming influence functions"
   },
   {
    "what": "the specification",
    "file": "data/texts/AXN-0700-text.md",
    "read": "§0–§11 whole; Appendix A by its opening"
   },
   {
    "what": "the briefs",
    "file": "rebuild/negative-of-the-negative/v09/BRIEF-NEW-ROW-AGGREGATOR.md, BRIEF-NEW-ROW-PROVENANCE.md, BRIEF.md",
    "read": "whole"
   }
  ],
  "cuts": [],
  "notes": "equal case; KO 27 sentences, all 51 field and 39 archive claims cited; P body 323 words; KO_B 14 sentences, 0 surviving at ≥ 0.95"
 },
 "field_by_origin": {
  "cards": [
   "B1",
   "B2",
   "B3",
   "B4",
   "B5",
   "B6"
  ],
  "own_pages": [
   "B7"
  ],
  "wikipedia": [
   "B8"
  ],
  "census_third_party": [],
  "unreachable": [
   "B4",
   "B5"
  ],
  "title_only": [
   "B4",
   "B5"
  ],
  "wikipedia_state": "no article: 'OLMoTrace' and 'OlmoTrace' are both absent (raw wikitext HTTP 404; rendered page HTTP 404 'Wikipedia does not have an article with this exact name', after one HTTP 429 refetched), no redirect, and a full-text search of article space for \"OLMoTrace\" returns 'There were no results matching the query.' (2026-10-29)",
  "sources": [
   {
    "id": "B1",
    "origin": "card",
    "role": "Overview card 1",
    "card": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "url": "https://allenai.org/blog/olmotrace",
    "kind": "Overview card (the operator's own blog post, 9 April 2025, 'Jiacheng Liu et al. - Ai2')",
    "files": {
     "B1.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B1.html",
      "sha256": "39e59dda3ba78e0bcf803c029242a75ae1e877da9383b35712bb1dc48ca2eeba"
     },
     "B1.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B1.txt",
      "sha256": "7d7b29ab8ba6c6e171e7b1cbc24f181c0c8d4a553f28af8e3af7ca6c2124864f"
     }
    },
    "fetched": {
     "at": "2026-10-29T17:07:13Z",
     "http": 200,
     "url": "https://allenai.org/blog/olmotrace",
     "raw": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B1.html",
     "raw_sha256": "39e59dda3ba78e0bcf803c029242a75ae1e877da9383b35712bb1dc48ca2eeba",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B1.txt"
     ]
    },
    "names_operator": "yes ('Ai2', 'Jiacheng Liu et al. - Ai2', '© The Allen Institute for Artificial Intelligence')",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "note": "URL found by WebSearch on the card's title and site, confirmed by the page's <title> (identical to the card title). The card's snippet ('Feature: OLMoTrace in Ai2 Playground traces language model outputs to multi-trillion-token training data in real time. Supported Models: OLMo 2 32B Instruct, OL...') is not in the served page: neither 'Feature:' nor 'Supported Models' occurs in B1.html, and the page's meta description is Ai2's generic one; the snippet's words are not the page's. Recorded, not resolved.",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 1
     }
    ],
    "card_title": "Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "card_snippet": "Feature: OLMoTrace in Ai2 Playground traces language model outputs to multi-trillion-token training data in real time. Supported Models: OLMo 2 32B Instruct, OL...",
    "card_note": null
   },
   {
    "id": "B2",
    "origin": "card",
    "role": "Overview card 2",
    "card": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "url": "https://arxiv.org/abs/2504.07096",
    "kind": "Overview card (the paper's abstract page, v2; the census's own page olmotrace--abs)",
    "files": {
     "B2.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B2.html",
      "sha256": "cd982680a1ca4c02a28ebd0cb3c7f7c1b7ed60cde75787fd3918c12f99154c22"
     },
     "B2.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B2.txt",
      "sha256": "836bf63640ea080a6becd9996270262dc568fcbc2a10ec960748ffa1515812f0"
     }
    },
    "fetched": {
     "at": "2026-10-29T17:07:17Z",
     "http": 200,
     "url": "https://arxiv.org/abs/2504.07096",
     "raw": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B2.html",
     "raw_sha256": "cd982680a1ca4c02a28ebd0cb3c7f7c1b7ed60cde75787fd3918c12f99154c22",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B2.txt"
     ],
     "census_copy": {
      "src": "olmotrace--abs",
      "sha256": "cd982680a1ca4c02a28ebd0cb3c7f7c1b7ed60cde75787fd3918c12f99154c22",
      "read": "2026-10-29",
      "http": "200",
      "agrees": true
     }
    },
    "names_operator": "yes (the 31 authors; the paper's affiliations are on the HTML version, B7)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "note": "URL from the card's snippet ('[2504.07096] OLMoTrace: …'). The fetch is byte-identical to the census's saved copy (sha256 cd982680…, read the same day), so the card and the census's own page are one page, carried once, as a card. The snippet's 'Scheduled maintenance: Wednesday, October 28, 7:00 to 12:00 EDT' is arXiv's site banner, not the paper's text, and was not on the page at this fetch.",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 2
     }
    ],
    "card_title": "OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "card_snippet": "[2504.07096] OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens Scheduled maintenance: Wednesday, October 28, 7:00 to 12:00 EDT (11:",
    "card_note": null
   },
   {
    "id": "B3",
    "origin": "card",
    "role": "Overview card 3",
    "card": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
    "url": "https://aclanthology.org/2025.acl-demo.18/",
    "kind": "Overview card (the paper's Anthology page, Proceedings of ACL 2025, System Demonstrations, with its PDF)",
    "files": {
     "B3.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3.html",
      "sha256": "71292b784f0dc598a840336f806fcf71cdd1ce9a332575a420315d36c1a9041d"
     },
     "B3.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3.txt",
      "sha256": "26daccd5e52cd634925647632ccd4f6b800c561b6bb7056d3eb5cd97c575f3cf"
     },
     "B3-pdf.pdf": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3-pdf.pdf",
      "sha256": "fcac45353235b662261103efb51083cd12e67c2b5b992b53eda0ceef5d13dd17"
     },
     "B3-pdf.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3-pdf.txt",
      "sha256": "f2fc53f52892169d4e7f5362d06f3c59fc46e29dda64b49bc707afcfa4cb4cbd"
     }
    },
    "fetched": {
     "at": "2026-10-29T17:07:19Z",
     "http": 200,
     "url": "https://aclanthology.org/2025.acl-demo.18/",
     "raw": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3.html",
     "raw_sha256": "71292b784f0dc598a840336f806fcf71cdd1ce9a332575a420315d36c1a9041d",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3.txt",
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B3-pdf.txt"
     ],
     "pdf": {
      "url": "https://aclanthology.org/2025.acl-demo.18.pdf",
      "at": "2026-10-29T17:09:20Z",
      "http": 200,
      "sha256": "fcac45353235b662261103efb51083cd12e67c2b5b992b53eda0ceef5d13dd17",
      "text": "pdftotext (poppler), default reading order"
     }
    },
    "names_operator": "yes (the authors; the PDF carries the affiliations)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "note": "URL found by WebSearch on the card's title and site (aclanthology.org/2025.acl-demo.18). The page's abstract is the card's snippet. The published paper (the page's PDF link) was fetched beside the page so the card's own paper could be read for its stated limit; its text is B3-pdf.txt.",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 3
     }
    ],
    "card_title": "Tracing Language Model Outputs Back to Trillions of Training Tokens",
    "card_snippet": "We present OLMOTRACE, the first system that traces the outputs of language models back to their full, multi-trillion-token training data in real time. OLMOTRACE...",
    "card_note": null
   },
   {
    "id": "B4",
    "origin": "card",
    "role": "Overview card 4",
    "card": "X — Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
    "url": null,
    "kind": "Overview card (a post by Ai2 on X; unreachable)",
    "files": {
     "B4-try1.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B4-try1.html",
      "sha256": "86472af42ddd3df14fb3846b3a437a006749633744b3764f19e25b26d7088778"
     },
     "B4-try2.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B4-try2.html",
      "sha256": "cf49e87cc38bcf452d79b701c68ef9797fcc5aa27fcac0205a928294aadd00e3"
     }
    },
    "fetched": {
     "unreachable": true,
     "tries": [
      {
       "at": "2026-10-29T17:07:24Z",
       "url": "https://x.com/allen_ai",
       "http": 200,
       "sha256": "86472af42ddd3df14fb3846b3a437a006749633744b3764f19e25b26d7088778",
       "result": "the profile's script shell (<title>Ai2 (@allen_ai) / X</title>); no post text; 'OlmoTrace' occurs nowhere in the response"
      },
      {
       "at": "2026-10-29T17:07:25Z",
       "url": "https://x.com/search?q=%22wired%20into%20OlmoTrace%22",
       "http": 200,
       "sha256": "cf49e87cc38bcf452d79b701c68ef9797fcc5aa27fcac0205a928294aadd00e3",
       "result": "the site's script shell (<title>X - The Everything App / X</title>); no results text"
      }
     ],
     "searches": [
      "WebSearch 'Ai2 x.com \"Olmo 3 is wired into OlmoTrace in the Ai2 Playground\"' (no x.com result)",
      "WebSearch '\"OlmoTrace\" \"Olmo 3-Think\" allen_ai status x.com trace' (no x.com result)",
      "WebSearch '\"wired into OlmoTrace\"' (no x.com result)"
     ],
     "note": "The post's status URL could not be found: three searches on the card's title returned no x.com status link, and x.com serves only a script shell without the post to an unauthenticated fetch. Recorded as unreachable; the card's title and snippet stand as its only text."
    },
    "names_operator": "yes ('Ai2 on X', in the snippet)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 4
     }
    ],
    "card_title": "Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
    "card_snippet": "Ai2 on X: \"Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just see its behavior—you can trace it. For example, you can ask Olmo 3-Think (32B...",
    "card_note": null
   },
   {
    "id": "B5",
    "origin": "card",
    "role": "Overview card 5",
    "card": "Techzine Global — Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
    "url": "https://www.techzine.eu/news/analytics/130296/ai2-introduces-olmotrace-a-peek-inside-llms/",
    "kind": "Overview card (unreachable: HTTP 403)",
    "files": {
     "B5-403.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B5-403.html",
      "sha256": "f261423d1da97890ff6bfccd31ff4ee0b74f69351cf75095fbe791cb16cbf5bd"
     }
    },
    "fetched": {
     "unreachable": true,
     "tries": [
      {
       "at": "2026-10-29T17:07:21Z",
       "http": 403,
       "sha256": "f261423d1da97890ff6bfccd31ff4ee0b74f69351cf75095fbe791cb16cbf5bd",
       "result": "'403 Forbidden' (103 bytes, saved as B5-403.html)"
      },
      {
       "at": "2026-10-29T17:07:47Z",
       "http": 403,
       "result": "the same 103 bytes (same sha256)"
      },
      {
       "at": "2026-10-29T17:07:48Z",
       "http": 403,
       "result": "the same 103 bytes (same sha256)"
      }
     ],
     "note": "URL found by WebSearch on the card's title and site. Three fetches with the brief's command (curl -sSL -m 30 -A \"Mozilla/5.0\") returned HTTP 403; a fetch with other request headers was refused by the workspace as bot-check circumvention and not pursued. The card's title and snippet stand as its only text."
    },
    "names_operator": "yes ('Ai2' in the title)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 5
     }
    ],
    "card_title": "Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
    "card_snippet": "After generating an answer from OLMo in the Ai2 Playground, users can activate the tool by clicking the “Show OLMoTrace” button below the output. The tool then ...",
    "card_note": null
   },
   {
    "id": "B6",
    "origin": "card",
    "role": "Overview card 6",
    "card": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "url": "https://theaieconomy.substack.com/p/meet-olmotrace-the-tool-that-shows",
    "kind": "Overview card (a newsletter post, Ken Yeung, 9 April 2025, with statements by Ai2's Liu, Farhadi, Lebrecht and Hajishirzi)",
    "files": {
     "B6.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B6.html",
      "sha256": "bf021478cec54cb8674a5919ddffc6ddbaad08640ab125d0c0bc96a1b7dabcaf"
     },
     "B6.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B6.txt",
      "sha256": "79b8019f592e2b12f4bf7d2aa2e49cb0dd9be3fa788abe45ac976fc918851eec"
     }
    },
    "fetched": {
     "at": "2026-10-29T17:07:23Z",
     "http": 200,
     "url": "https://theaieconomy.substack.com/p/meet-olmotrace-the-tool-that-shows",
     "raw": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B6.html",
     "raw_sha256": "bf021478cec54cb8674a5919ddffc6ddbaad08640ab125d0c0bc96a1b7dabcaf",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B6.txt"
     ]
    },
    "names_operator": "yes ('Ai2'; Jiacheng Liu, Ali Farhadi, Sophie Lebrecht, Hannaneh Hajishirzi named and quoted)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "note": "URL from the newsletter's own archive API (theaieconomy.substack.com/api/v1/archive?search=olmotrace: title 'Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info', 2025-04-09), confirmed by the page's text, which opens with the card's snippet.",
    "surfaced_by_T": [
     {
      "slug": "olmotrace-aio-20261029",
      "card": 6
     }
    ],
    "card_title": "Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "card_snippet": "Ken Yeung Apr 09, 2025 1 Ai2 has launched a tool to help developers answer a critical question: Where do AI models get their data? Called OLMoTrace, it's an ope...",
    "card_note": null
   },
   {
    "id": "B7",
    "origin": "own_page",
    "role": "the census's own page olmotrace--paper (not surfaced by T)",
    "card": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "url": "https://arxiv.org/html/2504.07096v2",
    "kind": "own page (the paper, v2, 8 July 2025), cited by the provenance census record; copied from the census's saved copy",
    "files": {
     "B7.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B7.html",
      "sha256": "b1b16ba583eeb0dbfdb07615ebda0ce56623b190b14c63115ff04a8a374e0bdd"
     },
     "B7.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B7.txt",
      "sha256": "0d848290949c755204a93f016a8c0484746d4f940b451e5e2489a1d854c403e7"
     }
    },
    "fetched": {
     "census_copy": true,
     "src": "olmotrace--paper",
     "read": "2026-10-29",
     "http": "200",
     "url": "https://arxiv.org/html/2504.07096v2",
     "raw": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B7.html",
     "raw_sha256": "b1b16ba583eeb0dbfdb07615ebda0ce56623b190b14c63115ff04a8a374e0bdd",
     "census_sha256": "b1b16ba583eeb0dbfdb07615ebda0ce56623b190b14c63115ff04a8a374e0bdd",
     "agrees": true,
     "copied_from": "/home/claude/fleet/provenance-erasure/data/provenance-census/sources/olmotrace--paper.html.gz (gunzipped) and olmotrace--paper.txt",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B7.txt"
     ]
    },
    "surfaced_by_T": [],
    "names_operator": "yes (the affiliations: Allen Institute for AI, University of Washington, UC Berkeley, Stanford University)",
    "names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, provenanceerasure, machinemediation, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
    "note": "Admitted beside the cards by the operator's ruling of 2026-10-29 ('as against their wikipedia articles if they exist, their own pages, etc.'); not surfaced by T, though it is the HTML version linked from card 2's page. The census's saved copy is intact (its sha256 matches the census's record), so it was copied and not refetched."
   },
   {
    "id": "B8",
    "origin": "wikipedia",
    "role": "the entity's Wikipedia article (none)",
    "card": "Wikipedia — OLMoTrace (no article)",
    "url": "https://en.wikipedia.org/w/index.php?title=OLMoTrace&action=raw",
    "kind": "Wikipedia (no article)",
    "files": {
     "B8.wiki": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8.wiki",
      "sha256": "400fcefc4db2f0f44126f924e05490ca0ceb15477d3b20d3dba8c36871072e29"
     },
     "B8.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8.html",
      "sha256": "92d530a8e9a8411fca35e6e3b8e10f2c8ae026db303fe423d578d5c4b2f22e81"
     },
     "B8-search.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8-search.html",
      "sha256": "3649e1b4ebe982d77b8e5dcf804774f72cd3bdfe1c79e06fc164bb5cfc6c3618"
     },
     "B8-variant.wiki": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8-variant.wiki",
      "sha256": "d3059a55262eb00fdf419f19f9cdffe19f9a33a91f63c747116357120f788979"
     },
     "B8-variant.html": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8-variant.html",
      "sha256": "35a2b545fe81f363bec407be04d98f41f91a2fadd58fdd7177dd68f203e25226"
     },
     "B8.txt": {
      "path": "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8.txt",
      "sha256": "e0353e047d8f2fa2f2fbce22950a4cb00a20940a12e0b89dbcc9c25017160d2b"
     }
    },
    "fetched": {
     "tries": [
      {
       "at": "2026-10-29T17:09:54Z",
       "url": "https://en.wikipedia.org/w/index.php?title=OLMoTrace&action=raw&redirect=no",
       "http": 404
      },
      {
       "at": "2026-10-29T17:09:54Z",
       "url": "https://en.wikipedia.org/wiki/OLMoTrace?redirect=no",
       "http": 429,
       "result": "rate-limited; refetched"
      },
      {
       "at": "2026-10-29T17:10:09Z",
       "url": "https://en.wikipedia.org/wiki/OLMoTrace?redirect=no",
       "http": 404,
       "result": "'Wikipedia does not have an article with this exact name.'"
      },
      {
       "at": "2026-10-29T17:09:55Z",
       "url": "https://en.wikipedia.org/w/index.php?title=OlmoTrace&action=raw&redirect=no",
       "http": 404
      },
      {
       "at": "2026-10-29T17:09:55Z",
       "url": "https://en.wikipedia.org/wiki/OlmoTrace?redirect=no",
       "http": 404
      },
      {
       "at": "2026-10-29T17:09:58Z",
       "url": "Special:Search \"OLMoTrace\" (article space)",
       "http": 429,
       "result": "rate-limited; refetched"
      },
      {
       "at": "2026-10-29T17:10:10Z",
       "url": "https://en.wikipedia.org/w/index.php?search=%22OLMoTrace%22&title=Special:Search&profile=advanced&fulltext=1&ns0=1",
       "http": 200,
       "result": "'There were no results matching the query.'"
      }
     ],
     "raw_sha256": "92d530a8e9a8411fca35e6e3b8e10f2c8ae026db303fe423d578d5c4b2f22e81",
     "text": [
      "datasets/negative-of-the-negative/v2/intake/olmotrace-20261029/B8.txt"
     ]
    },
    "surfaced_by_T": [],
    "names_operator": "no",
    "names_archive": "no",
    "note": "No article and no redirect for either spelling, and no article text containing the name."
   }
  ]
 }
}