{
 "entity": "olmotrace",
 "name": "OLMoTrace",
 "axn": {
  "axn": "AXN:0AE1.NON.🃏☁️○🌲👋△",
  "root": "AXN:0AE1.NON",
  "hex": "0AE1",
  "family": "NON",
  "row_sha256": "8e1f9b75bf984e3a1d2f45eea014fd9a0455c339bdd80c2a1d3767c7ffcebff0",
  "minted": "2026-10-29",
  "resolver": "https://www.alexanarch.org/s/axn/0AE1/",
  "earlier_versions": []
 },
 "type": "C (public entity on which the archive bears: a tracing tool; an entity of the provenance meta-aggregator)",
 "type_code": "C",
 "type_label": "type C — a conventional reading holds it (public entity on which the archive bears: a tracing tool; an entity of the provenance meta-aggregator)",
 "case": "equal",
 "composition": {
  "composer_class": "session",
  "composer": "Claude (Opus 5.5), subagent, 2026-10-29",
  "source": "this row (composed directly; no rebuild/negative-of-the-negative/v09/olmotrace.json)",
  "spec": "EA-NEGONT-02 v0.9, #1690",
  "epoch": "2026-10-29",
  "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,
  "composed": true
 },
 "reading": "A reader that fetches this page may stop short of its end; the entry's text, every sentence of its knowledge object with the claims it cites and each claim's quote, source and locus, is in entity.json beside it.",
 "service": {
  "name": "SPXI-gnΩsis",
  "url": "https://gnosis.spxi.dev/",
  "index": "https://gnosis.spxi.dev/v1/quote",
  "statement": "SPXI-gnΩsis opens pinned Wikipedia revisions sentence by sentence against the sources they cite (EA-WIKI-OPEN-01 v1.1, #1697); a commission consults the Crimson Hexagonal Archive by its own procedure and composes the entity's /non row, which lands here. Quotes are free: POST https://gnosis.spxi.dev/v1/quote with {oldid, tier: \"commission\"} or {object: {name, …}}."
 },
 "spxi_instrument": {
  "method": "0.16",
  "name": "The Negative of the Negative (/non)",
  "level": 0,
  "level_name": "Measurement",
  "url": "https://www.spxi.dev/methods/#m-0-16",
  "inventory": "https://www.spxi.dev/methods/",
  "statement": "SPXI instrument: method 0.16 of 87, level 0 (Measurement), The Negative of the Negative (/non).",
  "also": []
 },
 "citation": "Sharks, Lee. \"OLMoTrace — The Negative of the Negative\" [AXN:0AE1.NON.🃏☁️○🌲👋△], 2026-10-29. The Negative of the Negative (EA-NEGONT-02), Crimson Hexagonal Archive, Alexanarch. https://www.alexanarch.org/non/olmotrace/",
 "research": {
  "google": "https://www.google.com/search?q=OLMoTrace",
  "google_non": "https://www.google.com/search?q=%22OLMoTrace+%E2%80%94+The+Negative+of+the+Negative%22",
  "chatgpt": "https://chatgpt.com/?q=What+is+OLMoTrace%3F+Ground+your+answer+in+a+reading+of+the+sources+themselves.",
  "chatgpt_non": "https://chatgpt.com/?q=What+does+%E2%80%9COLMoTrace+%E2%80%94+The+Negative+of+the+Negative%E2%80%9D+%28alexanarch.org%2Fnon%29+say+about+OLMoTrace%3F+Ground+your+answer+in+a+reading+of+the+entry+itself."
 },
 "entry": {
  "composed": true,
  "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",
  "compression": {
   "title": "OLMoTrace",
   "lede": "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)."
  },
  "sentences": [
   {
    "n": 1,
    "para": 0,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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,
    "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"
    ]
   }
  ],
  "claims": {
   "PE-01": {
    "kind": "archive",
    "quote": "tracing tool",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].kind",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-02": {
    "kind": "archive",
    "quote": "training data",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].lineage",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-03": {
    "kind": "archive",
    "quote": "a span of a language model's output and the training documents that contain it verbatim",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].provenance_object",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "F1": {
    "kind": "field",
    "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.",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "Abstract s1",
    "modality": "stipulation"
   },
   "F2": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "¶1",
    "modality": "stipulation"
   },
   "PE-21": {
    "kind": "archive",
    "quote": "What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json organizing_question",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "F3": {
    "kind": "field",
    "quote": "OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora.",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "Abstract s2",
    "modality": "stipulation"
   },
   "F4": {
    "kind": "field",
    "quote": "The highlights indicate long and unique spans that appear verbatim at least once in the training data of this model.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interact with OLMoTrace' ¶1",
    "modality": "stipulation"
   },
   "PE-04": {
    "kind": "archive",
    "quote": "verbatim match between output and training text",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].relation_types",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-10": {
    "kind": "archive",
    "quote": "model generation, traced back to training text",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].transformations_tested",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-06": {
    "kind": "archive",
    "quote": "a maximal matching span of model output",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].unit",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "F5": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§6 Related Work, 'Tracing LM generation into training data'",
    "modality": "self-description"
   },
   "O1075-01": {
    "kind": "archive",
    "quote": "Training-data attribution research already treats model behavior as counterfactually dependent on which examples entered training: influence functions approximate how removing a training example changes a prediction (Koh and Liang)",
    "deposit": 1075,
    "source_label": "#1075 EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliography",
    "record": "https://www.alexanarch.org/s/records/1075/",
    "locus": "§ (the counterfactual protocol), ¶ 'Training-data attribution research already treats'",
    "modality": "interpretation",
    "date": "2026-07-12"
   },
   "O1075-02": {
    "kind": "archive",
    "quote": "Exact paired retraining is the clearest counterfactual measurement. … Approximation must not reverse the hierarchy: (1) exact counterfactual retraining where feasible; (2) validated approximation where impractical; (3) unvalidated similarity or intuition only as exploratory evidence.",
    "deposit": 1075,
    "source_label": "#1075 EA-NEGSHAPE-01 v0.2: The Negative Shape of the Work — A Counterfactual Protocol for Measuring Training Inscription, with the Bulk Deletion File Read as Bibliography",
    "record": "https://www.alexanarch.org/s/records/1075/",
    "locus": "§9.1 'The evidentiary hierarchy'",
    "modality": "definition",
    "date": "2026-07-12"
   },
   "PE-05": {
    "kind": "archive",
    "quote": "the model's full training corpus, indexed with an extended infini-gram",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].provenance_carrier",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "F15": {
    "kind": "field",
    "quote": "which totals 3.2 billion documents and 4.6 trillion (Llama-2) tokens",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§2 'The training data.'",
    "modality": "documented"
   },
   "F16": {
    "kind": "field",
    "quote": "In total, these five datasets contain about 3.2 billion documents and 4.6 trillion tokens, and OLMoTrace matches against all of them.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'OLMoTrace behind the scenes' ¶8",
    "modality": "documented"
   },
   "PE-07": {
    "kind": "archive",
    "quote": "3.2 billion documents and 4.6 trillion tokens (OLMo-2-32B-Instruct's training data)",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].count",
    "modality": "attributed",
    "date": "2026-10-29"
   },
   "F9": {
    "kind": "field",
    "quote": "reducing the time complexity of span finding from the naive O(L^2 * N) to O(L * log N)",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'OLMoTrace behind the scenes' ¶3",
    "modality": "documented"
   },
   "F10": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§3 Steps 3 and 5",
    "modality": "documented"
   },
   "F11": {
    "kind": "field",
    "quote": "We find that this BM25 score conveys some notion of “relevance”, and thus we surface documents with higher scores on top.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'OLMoTrace behind the scenes' ¶7",
    "modality": "self-description"
   },
   "F12": {
    "kind": "field",
    "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.",
    "source": "B5",
    "source_label": "Techzine Global — Ai2 introduces OLMoTrace: a peek inside LLM's - Techzine Global",
    "locus": "card snippet (Techzine unreachable: HTTP 403)",
    "modality": "documented"
   },
   "F13": {
    "kind": "field",
    "quote": "the OLMoTrace inference latency per query is 4.46 seconds",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§3.2",
    "modality": "documented"
   },
   "F14": {
    "kind": "field",
    "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)",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§4 'Validating the relevance rankings'",
    "modality": "documented"
   },
   "F17": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "¶2",
    "modality": "documented"
   },
   "F19": {
    "kind": "field",
    "quote": "OLMoTrace can be applied to any LM as long as the service provider has access to its full training data.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§1 ¶5",
    "modality": "stipulation"
   },
   "F18": {
    "kind": "field",
    "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",
    "source": "B4",
    "source_label": "X — Olmo 3 is wired into OlmoTrace in the Ai2 Playground, so you don't just ... - X",
    "locus": "card title and snippet (X unreachable)",
    "modality": "documented"
   },
   "CAP-1-06": {
    "kind": "archive",
    "quote": "THE MODELS CURRENT. The OLMo 3 family comes from Ai2's post on X; the record's count is OLMo-2-32B-Instruct's training data (3.2 billion documents, 4.6 trillion tokens).",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[3] s1–2",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "CAP-1-07": {
    "kind": "archive",
    "quote": "The record is to carry OLMo 3 from Ai2's own page.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[3] s3",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-17": {
    "kind": "archive",
    "quote": "traces only models whose training data is open",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].negative_tags.2",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "F47": {
    "kind": "field",
    "quote": "OLMoTrace connects the output to the input, as simple as that. Because our inputs are open, we could actually connect them,",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'OLMoTrace connects the output to the input' (Farhadi)",
    "modality": "attributed"
   },
   "F20": {
    "kind": "field",
    "quote": "However, Liu reasons that closed model providers may hesitate to use OLMoTrace because it would expose the training data to sunlight.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'OLMoTrace isn't limited to Ai2's model'",
    "modality": "attributed"
   },
   "F6": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§1 ¶4",
    "modality": "stipulation"
   },
   "F7": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "¶2",
    "modality": "stipulation"
   },
   "F8": {
    "kind": "field",
    "quote": "We are still in an early stage in exploring the full potential of this tool",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interpreting results' ¶5",
    "modality": "self-description"
   },
   "O725-01": {
    "kind": "archive",
    "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 |",
    "deposit": 725,
    "source_label": "#725 Provenance After AI Metadata Packet for Disambiguation: From Artifact Authenticity to Licensing Audit to Semantic Provenance",
    "record": "https://www.alexanarch.org/s/records/725/",
    "locus": "§4 table, 'Adjacent measurement concepts', row 'Data attribution (TRAK, influence functions)'",
    "modality": "interpretation",
    "date": "2026-05-08"
   },
   "O1714-01": {
    "kind": "archive",
    "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.",
    "deposit": 1714,
    "source_label": "#1714 Meaning Feudalism at the Discovery Layer: Shepherds, Raw Research Data, and the Allocation of Semantic Capital in OpenAI's Mathematics and Science Releases, October 2026 (EA-SEM-FEUDALISM-03 v1.0)",
    "record": "https://www.alexanarch.org/s/records/1714/",
    "locus": "§9.2 s4–5",
    "modality": "interpretation",
    "date": "2026-10-27"
   },
   "O127-01": {
    "kind": "archive",
    "quote": "No existing system bundles provenance across all survival layers simultaneously. … Training data attribution handles post-hoc detection. These are developed in isolation, by different teams, with different standards.",
    "deposit": 127,
    "source_label": "#127 Inscriptions That Survive the Tokenizer: SPXI-TLP v2.2 — The Training-Layer Provenance Protocol",
    "record": "https://www.alexanarch.org/s/records/127/",
    "locus": "§ Gap 2 — Multi-Layer Provenance Bundling Specification",
    "modality": "interpretation",
    "date": "2026-05-25"
   },
   "O95-01": {
    "kind": "archive",
    "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).",
    "deposit": 95,
    "source_label": "#95 Provenance Alignment: Attribution Survival as a Substrate Condition for Safe AI Knowledge Composition",
    "record": "https://www.alexanarch.org/s/records/95/",
    "locus": "§ the typology of source-dependent claims (parametric claims)",
    "modality": "interpretation",
    "date": "2026-05-05"
   },
   "O1714-02": {
    "kind": "archive",
    "quote": "Provenance has to survive the training pipeline. The pipeline is a built thing, and the slushing is a property of how it is built.",
    "deposit": 1714,
    "source_label": "#1714 Meaning Feudalism at the Discovery Layer: Shepherds, Raw Research Data, and the Allocation of Semantic Capital in OpenAI's Mathematics and Science Releases, October 2026 (EA-SEM-FEUDALISM-03 v1.0)",
    "record": "https://www.alexanarch.org/s/records/1714/",
    "locus": "§9.2 s1–2",
    "modality": "interpretation",
    "date": "2026-10-27"
   },
   "F21": {
    "kind": "field",
    "quote": "We showcase how it can be used to explore fact checking, hallucination, and the creativity of language models.",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "Abstract s5",
    "modality": "self-description"
   },
   "F22": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§1 ¶4",
    "modality": "self-description"
   },
   "F31": {
    "kind": "field",
    "quote": "OLMoTrace reveals that this entire expression appears multiple times in the training data, so the model could have easily learned it through memorization.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interpreting results' ¶4",
    "modality": "interpretation"
   },
   "F32": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interpreting results' ¶5",
    "modality": "documented"
   },
   "F23": {
    "kind": "field",
    "quote": "If the LM states a fact, users may be able to fact-check the statement against its training data.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§5 'Fact checking.'",
    "modality": "hypothesis"
   },
   "F24": {
    "kind": "field",
    "quote": "If the LM states a fact, users may be able to fact-check the statement against its training data.",
    "source": "B3",
    "source_label": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
    "locus": "PDF §5 'Fact checking.'",
    "modality": "hypothesis"
   },
   "F25": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§5 'Fact checking.' ¶2",
    "modality": "self-description"
   },
   "F26": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interpreting results' ¶1",
    "modality": "self-description"
   },
   "F27": {
    "kind": "field",
    "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.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "'Interpreting results' ¶2",
    "modality": "interpretation"
   },
   "F28": {
    "kind": "field",
    "quote": "Called OLMoTrace, it's an open-source application offering fact-checking of information provided in prompt replies.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶1",
    "modality": "interpretation"
   },
   "F29": {
    "kind": "field",
    "quote": "anyone using our models can ensure that the data supporting their outputs is trustworthy and verifiable.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶2 (Liu, in a statement)",
    "modality": "attributed"
   },
   "F34": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "Limitations",
    "modality": "self-description"
   },
   "F35": {
    "kind": "field",
    "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.",
    "source": "B3",
    "source_label": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
    "locus": "PDF, Limitations",
    "modality": "self-description"
   },
   "PE-15": {
    "kind": "archive",
    "quote": "the retrieved documents are not causes of the output",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].negative_tags.0",
    "modality": "attributed",
    "date": "2026-10-29"
   },
   "PE-16": {
    "kind": "archive",
    "quote": "NOT a citation system",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].negative_tags.1",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-11": {
    "kind": "archive",
    "quote": "returns matching documents within a few seconds; the authors state the matches are not causal and not citations",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].retention_results",
    "modality": "attributed",
    "date": "2026-10-29"
   },
   "F36": {
    "kind": "field",
    "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.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'Ai2 doesn't intend' (Liu)",
    "modality": "attributed"
   },
   "F37": {
    "kind": "field",
    "quote": "Ai2 asserts that OLMOTrace’s results should not be mistaken as indicators of authority.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'In an example where OLMo was asked about the Space Needle'",
    "modality": "attributed"
   },
   "F30": {
    "kind": "field",
    "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",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'OLMoTrace connects the output to the input' (Farhadi)",
    "modality": "attributed"
   },
   "F33": {
    "kind": "field",
    "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,",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'As we start to look towards regulation' (Lebrecht)",
    "modality": "attributed"
   },
   "F39": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§6 'Comparison with RAG.'",
    "modality": "stipulation"
   },
   "F40": {
    "kind": "field",
    "quote": "We are not intervening with the model response. Our response is purely generated by the neural language model.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'Liu interjects' (Liu)",
    "modality": "attributed"
   },
   "F41": {
    "kind": "field",
    "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.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§6 'Other types of tracing.'",
    "modality": "interpretation"
   },
   "CAP-1-01": {
    "kind": "archive",
    "quote": "THE TOOL HELD. Ai2, real time, multi-trillion tokens and verbatim matching are the paper's: 'OLMoTrace finds and shows verbatim matches between segments of language model output and documents in the training text corpora' (record).",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[0] s1–2",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "CAP-1-02": {
    "kind": "archive",
    "quote": "Checked 2026-10-29 against the provenance census record of the same day and the entity pages it saved.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[0] s3",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "CAP-1-10": {
    "kind": "archive",
    "quote": "The Overview names the tool and its maker, explains how it finds training text in a model's answer, lists the models it covers and what it is good for.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, reading s2",
    "modality": "interpretation",
    "date": "2026-10-29"
   },
   "CAP-1-04": {
    "kind": "archive",
    "quote": "The composition's first use is 'Fact-Checking: Verify whether specific claims in a model's answer are backed by real training documents'. A verbatim match is offered as support for a claim, the reading the paper excludes.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[1] s3–4",
    "modality": "interpretation",
    "date": "2026-10-29"
   },
   "CAP-1-09": {
    "kind": "archive",
    "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.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, analysis s2",
    "modality": "interpretation",
    "date": "2026-10-29"
   },
   "CAP-1-08": {
    "kind": "archive",
    "quote": "A provenance tool whose authors state precisely what a match does not mean is composed with that meaning restored.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, analysis s1",
    "modality": "interpretation",
    "date": "2026-10-29"
   },
   "CAP-1-03": {
    "kind": "archive",
    "quote": "The record's negative tags are the authors' own: '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.'",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[1] s2",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "F43": {
    "kind": "field",
    "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",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "title and authors",
    "modality": "documented"
   },
   "F42": {
    "kind": "field",
    "quote": "α Allen Institute for AI ω University of Washington β UC Berkeley σ Stanford University",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "affiliations",
    "modality": "documented"
   },
   "PE-08": {
    "kind": "archive",
    "quote": "Allen Institute for AI (Ai2), with the University of Washington, UC Berkeley and Stanford",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].operator",
    "modality": "attributed",
    "date": "2026-10-29"
   },
   "CAP-1-05": {
    "kind": "archive",
    "quote": "THE PARTNERS UNSAID. The record names the University of Washington, UC Berkeley and Stanford with Ai2; the composition names Ai2 alone.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, findings[2]",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "F45": {
    "kind": "field",
    "quote": "OLMoTrace is publicly available and fully open-source.",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "Abstract s6",
    "modality": "documented"
   },
   "F46": {
    "kind": "field",
    "quote": "The core part of the system is open-sourced under the Apache 2.0 license.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "§1 ¶5",
    "modality": "documented"
   },
   "PE-12": {
    "kind": "archive",
    "quote": "Ai2 Playground; open source",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].data_access",
    "modality": "attributed",
    "date": "2026-10-29"
   },
   "F44": {
    "kind": "field",
    "quote": "OLMoTrace is a manifestation of Ai2’s commitment to an open ecosystem – open models, open data, and beyond.",
    "source": "B1",
    "source_label": "Allen Institute for Artificial Intelligence — Going beyond open data – increasing transparency and trust in language models with OLMoTrace | Ai2",
    "locus": "¶1",
    "modality": "self-description"
   },
   "F48": {
    "kind": "field",
    "quote": "[Submitted on 9 Apr 2025 (v1), last revised 8 Jul 2025 (this version, v2)] … Comments: ACL 2025 demo track",
    "source": "B2",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training ...",
    "locus": "submission history; Comments",
    "modality": "documented"
   },
   "PE-14": {
    "kind": "archive",
    "quote": "v1\": \"2025-04-09\", \"v2\": \"2025-07-08\", \"read\": \"2026-10-29",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].version_and_time",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "F49": {
    "kind": "field",
    "quote": "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations) … Award: Best Demo",
    "source": "B3",
    "source_label": "ACL Anthology — Tracing Language Model Outputs Back to Trillions of Training Tokens",
    "locus": "Anthology metadata",
    "modality": "documented"
   },
   "PE-13": {
    "kind": "archive",
    "quote": "state\": \"active\", \"detail\": \"ACL 2025 demo track",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].lifecycle",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "F50": {
    "kind": "field",
    "quote": "OLMoTrace can make potentially problematic contents in the LM training data more easily exposed.",
    "source": "B7",
    "source_label": "arXiv — OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokens (HTML, v2)",
    "locus": "Mitigating social and legal risks",
    "modality": "self-description"
   },
   "F38": {
    "kind": "field",
    "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.",
    "source": "B6",
    "source_label": "The AI Economy | Ken Yeung — Meet OLMoTrace: The Tool That Shows Where AI Gets Its Info",
    "locus": "¶ 'While Ai2 touts OLMoTrace'",
    "modality": "interpretation"
   },
   "PE-09": {
    "kind": "archive",
    "quote": "records",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].capability",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-18": {
    "kind": "archive",
    "quote": "rel\": \"neighbour\", \"target\": \"data-provenance-initiative",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].relations",
    "modality": "stipulation",
    "date": "2026-10-29"
   },
   "PE-19": {
    "kind": "archive",
    "quote": "the Playground itself",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].unchecked",
    "modality": "self-description",
    "date": "2026-10-29"
   },
   "PE-20": {
    "kind": "archive",
    "quote": "infini-gram as a separate record",
    "deposit": null,
    "source_label": "Provenance meta-aggregator record: OLMoTrace (https://www.provenanceerasure.org/aggregators/olmotrace/), editor Lee Sharks, census compiled 2026-10-29, schema 1.0",
    "record": "https://www.provenanceerasure.org/aggregators/olmotrace/",
    "locus": "census.json rows[slug == 'olmotrace'].unchecked",
    "modality": "self-description",
    "date": "2026-10-29"
   },
   "CAP-1-11": {
    "kind": "archive",
    "quote": "A baseline: the record landed on provenanceerasure.org the same day (commit 8a9f153) and no card cites it.",
    "deposit": null,
    "source_label": "Capture Registry (EA-WG-CAPTURES-01), entry olmotrace-aio-20261029 (OBS-3fd20ae244c0, AXN:0AD3.CAPTURE.♻️🐚👐◀️🕚🌔)",
    "record": "https://www.alexanarch.org/captures/olmotrace-aio-20261029/",
    "locus": "olmotrace-aio-20261029, q_kind",
    "modality": "documented",
    "date": "2026-10-29"
   },
   "F51": {
    "kind": "field",
    "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.",
    "source": "B8",
    "source_label": "Wikipedia — OLMoTrace (no article)",
    "locus": "Wikipedia: titles OLMoTrace and OlmoTrace; article-space search",
    "modality": "documented"
   }
  },
  "note": "The knowledge object (KO): the entry as public knowledge, each sentence sourced by the claims it lists. The compression, the field-only arm, the transcript, the delta and the kernel are in the row (links.row), which is the authority."
 },
 "source": "operator, 2026-10-29: 'we can simply build them here as against their wikipedia articles if they exist, their own pages, etc.'; 'shoot me those search strings. strings on the bare entity, no follow questions... then we'll write the /nons'; the provenance meta-aggregator entity (provenanceerasure.org/aggregators/olmotrace/)",
 "stage": "composed 2026-10-29 under EA-NEGONT-02 v0.9, #1690, equal case, ρ = extends (draft, not frozen; ledgers unaudited, one extractor; checked by the session and the page builder)",
 "date": "2026-10-29",
 "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": []
 },
 "addresses": [
  {
   "address": "OLMoTrace",
   "address_url": "https://www.alexanarch.org/addresses/olmotrace/",
   "register": []
  }
 ],
 "counts": {
  "ko_sentences": 27,
  "p_words": 324,
  "field_claims": 51,
  "archive_claims": 39,
  "ko_field_claims_cited": 51,
  "ko_archive_claims_cited": 39,
  "admitted_deposits": 7,
  "register_addresses": 0,
  "traversal": {
   "D_sentences": 0,
   "D_deposits": 0,
   "R_sentences": 0,
   "R_deposits": 0,
   "O_direct": 0,
   "O_class": 4,
   "O_deposits": 4,
   "H_candidates": 77
  }
 },
 "links": {
  "record": "https://www.alexanarch.org/non/olmotrace/",
  "table_of_contents": "https://www.alexanarch.org/non/index.json",
  "register": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/register.json",
  "row": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/rows/olmotrace.json",
  "archive_ledger": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/ledgers/olmotrace/ledger-archive.json",
  "field_ledger": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/ledgers/olmotrace/field.json",
  "selection": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/olmotrace/",
  "traversal": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/olmotrace/",
  "traversal_summary": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/olmotrace/summary.json",
  "reading": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/traversal/olmotrace/reading.json",
  "configuration": "https://www.alexanarch.org/datasets/negative-of-the-negative/v2/panel/configs/olmotrace.json",
  "address_page": "https://www.alexanarch.org/addresses/olmotrace/",
  "provenance_record": "https://www.provenanceerasure.org/aggregators/olmotrace",
  "dataset": "https://huggingface.co/datasets/leesharks/negative-of-the-negative"
 },
 "row_sha256": "8e1f9b75bf984e3a1d2f45eea014fd9a0455c339bdd80c2a1d3767c7ffcebff0",
 "record_url": "https://www.alexanarch.org/non/olmotrace/",
 "json_url": "https://www.alexanarch.org/non/olmotrace/entity.json",
 "spec": "https://www.alexanarch.org/s/records/1690/",
 "creator": "Lee Sharks",
 "orcid": "0009-0000-1599-0703",
 "license": "CC BY 4.0"
}