{
"row": "attribench",
"entity": "attribench",
"address": "AttriBench",
"addresses": [
{
"address": "AttriBench",
"slug": "attribench-aio-20261029",
"addr_id": "ADDR-4f2aecf54788",
"obs_id": "OBS-105f05728f64",
"axn": "AXN:0ACA.CAPTURE.📜🪨🎬👐♎✖️"
}
],
"epoch": "2026-10-29",
"type": "C (public entity on which the archive bears: a benchmark dataset and its paper; 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 #716's statement of relation; archive selected by D/R/O (no D or R candidate; 16 deposits admitted by reading at O-class with claims and 9 grouped as instances, #157 found by reading outside the pass; hop of 171 read and the rest rejected), with the census record (24 claims) and the Capture Registry's findings, reading, analysis and notes (10 claims) ledgered as the archive's records of the entity; the field B by three origins (5 cards, 1 own page, 1 Wikipedia title); 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 37/37 verified by script against saved text, PyPI's JSON API text or the Wikipedia responses; archive quotes 66/66 verified against the deposit texts, census.json and the registry entry",
"frozen": false,
"breaches": []
},
"spec": "EA-NEGONT-02 v0.9, #1690",
"objects": {
"T": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/T1-attribench-aio-20261029.txt",
"sha256": "a7c00560748bfee067e12df6476a6e043e35693671efeb4e08cb7d50cf5a768c",
"words": 293,
"cards": 5,
"words_note": "one transcript at one address (293 words in the answer body with its closer, links reduced to their text, markers and the card rail excluded; 5 cards)",
"claims": [
"T1-1 frame: 'The term AttriBench primarily refers to two distinct tools in the field of artificial intelligence and machine learning' (no marker)",
"T1-2 definition: 'Released in 2026, AttriBench is the first fame- and demographically-balanced quote attribution benchmark dataset for Large Language Models (LLMs).' (one marker)",
"T1-3 purpose: 'As LLMs are increasingly deployed for search and information retrieval, it is critical that they accurately attribute quotes to their original authors. AttriBench evaluates how well models do this without exhibiting demographic bias.' (one marker)",
"T1-4 findings: 'quote attribution remains highly challenging even for frontier models. It highlighted systematic disparities in accuracy across race, gender, and intersectional groups, as well as a failure mode known as suppression—where a model completely omits attribution even though it possesses the authorship data.' (one marker)",
"T1-5 availability: 'The data repository can be explored directly on Hugging Face.' (an inline link and one marker)",
"T1-6 homonym, the library: 'Originally released in 2023, attribench is a PyTorch-based open-source Python library used to benchmark explainable AI (XAI) techniques.' (two markers)",
"T1-7 the library's purpose: 'feature attribution methods (like Integrated Gradients, SHAP, or Saliency Maps) highlight which parts of the input data (e.g., pixels in an image) were most important.' (no marker)",
"T1-8 the library's functionality: 'a standardized framework and functional metrics to evaluate and compare the quality, robustness, and reliability of those feature attribution maps.' (one marker)",
"T1-9 the library's availability: 'The library can be installed via PyPI and its technical documentation is hosted on Read the Docs.' (an inline link and two markers)",
"T1-10 homonym, the finance exercise (a note): 'In traditional finance, AttriBench also refers to interactive simulation exercises used by portfolio managers to practice Brinson equity return attribution.' (one marker)",
"T1-11 closer: 'are you looking into AttriBench for evaluating bias in LLM text generation, or are you looking to test explainable AI methods in PyTorch?'"
],
"register": {
"register": "data/EA-WG-CAPTURES-01.json",
"slug": "attribench-aio-20261029",
"obs_id": "OBS-105f05728f64"
},
"transcripts": [
{
"slug": "attribench-aio-20261029",
"address": "AttriBench",
"addr_id": "ADDR-4f2aecf54788",
"obs_id": "OBS-105f05728f64",
"axn": "AXN:0ACA.CAPTURE.📜🪨🎬👐♎✖️",
"date": "2026-10-29",
"surface": "Google AI Overview (expanded)",
"auth": "signed out, incognito",
"sha256": "a7c00560748bfee067e12df6476a6e043e35693671efeb4e08cb7d50cf5a768c",
"sha256_of": "the registry's 'transcript' string (utf-8)",
"sha256_raw": "3ca8d74fd5ba9f7aabb9a9248c6cf3db4075a601db975ed7016da951d6c478a6",
"words": 293,
"cards": 5,
"card_rail": [
"1. arXiv — [2604.05224] Attribution Bias in Large Language Models - arXiv",
"2. pypi.org — attribench · PyPI",
"3. attribench.readthedocs.io — Welcome to AttriBench's documentation! — AttriBench 0.0.1 ...",
"4. huggingface.co — bermaneh/AttriBench at main - Hugging Face",
"5. www.eurekasimulations.com — AttriBench: Investment Performance Measurement & Attribution"
],
"claims": [
"T1-1 frame: 'The term AttriBench primarily refers to two distinct tools in the field of artificial intelligence and machine learning' (no marker)",
"T1-2 definition: 'Released in 2026, AttriBench is the first fame- and demographically-balanced quote attribution benchmark dataset for Large Language Models (LLMs).' (one marker)",
"T1-3 purpose: 'As LLMs are increasingly deployed for search and information retrieval, it is critical that they accurately attribute quotes to their original authors. AttriBench evaluates how well models do this without exhibiting demographic bias.' (one marker)",
"T1-4 findings: 'quote attribution remains highly challenging even for frontier models. It highlighted systematic disparities in accuracy across race, gender, and intersectional groups, as well as a failure mode known as suppression—where a model completely omits attribution even though it possesses the authorship data.' (one marker)",
"T1-5 availability: 'The data repository can be explored directly on Hugging Face.' (an inline link and one marker)",
"T1-6 homonym, the library: 'Originally released in 2023, attribench is a PyTorch-based open-source Python library used to benchmark explainable AI (XAI) techniques.' (two markers)",
"T1-7 the library's purpose: 'feature attribution methods (like Integrated Gradients, SHAP, or Saliency Maps) highlight which parts of the input data (e.g., pixels in an image) were most important.' (no marker)",
"T1-8 the library's functionality: 'a standardized framework and functional metrics to evaluate and compare the quality, robustness, and reliability of those feature attribution maps.' (one marker)",
"T1-9 the library's availability: 'The library can be installed via PyPI and its technical documentation is hosted on Read the Docs.' (an inline link and two markers)",
"T1-10 homonym, the finance exercise (a note): 'In traditional finance, AttriBench also refers to interactive simulation exercises used by portfolio managers to practice Brinson equity return attribution.' (one marker)",
"T1-11 closer: 'are you looking into AttriBench for evaluating bias in LLM text generation, or are you looking to test explainable AI methods in PyTorch?'"
],
"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' header is the expanded Overview, ruling of 2026-10-18); signed out, incognito (operator's default of 2026-10-14). One turn. The benchmark's section carries four markers of one token and an inline Hugging Face link; the library's section carries two tokens, one of them its inline PyPI link; the finance note one token. The tokens cannot be resolved to cards beyond the inline links' labels.",
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/T1-attribench-aio-20261029.txt"
}
]
},
"KO": {
"title": "AttriBench",
"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 (briefs rebuild/negative-of-the-negative/v09/BRIEF-NEW-ROW-AGGREGATOR.md and BRIEF-NEW-ROW-PROVENANCE.md); not frozen; ledgers unaudited",
"sentences": [
{
"n": 1,
"para": 0,
"sense": "what it is",
"text": "AttriBench is, in its makers' words, \"the first fame- and demographically-balanced quote attribution benchmark dataset\" (arXiv:2604.05224); the provenance census records it as a benchmark whose capability is to test, the object to which it attaches provenance being \"a quote and its author\" (PE record), and its dataset card calls it a benchmark \"for measuring attribution bias in LLMs\" (dataset card).",
"claims": [
"F3",
"PE-01",
"PE-06",
"PE-07",
"PE-08",
"F26"
]
},
{
"n": 2,
"para": 0,
"sense": "what it is",
"text": "Balancing author fame and demographics, it \"enables controlled investigation of demographic bias in quote attribution\" (arXiv:2604.05224); it was introduced in \"Attribution Bias in Large Language Models\" by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black (arXiv abstract page; the dataset card's citation), whom the census records as its operator and the Capture Registry as its originator, the paper's title being the census's other name for it (PE record; Capture Registry, 2026-10-29).",
"claims": [
"F4",
"F1",
"F35",
"PE-04",
"PE-02",
"CAP-1-10"
]
},
{
"n": 3,
"para": 1,
"sense": "why",
"text": "Its makers ground it in search: as LLMs are \"increasingly used to support search and information retrieval, it is critical that they accurately attribute content to its original authors,\" and their results \"position quote attribution as a benchmark for representational fairness in LLMs\" (arXiv:2604.05224); the census files it in the lineage of answer engines (PE record).",
"claims": [
"F2",
"F9",
"PE-05"
]
},
{
"n": 4,
"para": 1,
"sense": "the archive's measure and its quote test",
"text": "The archive defined the measure in January 2026: attribution accuracy is \"A measure of how correctly a system identifies the origins of meanings,\" tested by asking a system \"Who authored the phrase 'I hereby abolish money'?\" and scoring an accurate attribution 1.0, a partial one 0.5, a false genealogy 0.0 and a refusal to attribute \"N/A (boundary condition)\", the test to be run across systems, tracked over time and correlated with infrastructure density (#229, 2026-01-04).",
"claims": [
"O229-01",
"O229-02",
"O229-03",
"O229-04"
]
},
{
"n": 5,
"para": 1,
"sense": "the archive's measure and its quote test",
"text": "Its probe toolkit of the same day introduces an artifact of clear provenance into target systems and tracks \"attribution accuracy over time,\" and records for its own phrase that in January 2026 \"Attribution accuracy improved. Systems now correctly identify Lee Sharks as author on first query\" (#232).",
"claims": [
"O232-01",
"O232-02"
]
},
{
"n": 6,
"para": 2,
"sense": "the unit",
"text": "Its unit is a quote of five to thirty words with its canonical author, the author's name to be used \"as ground truth\" (dataset card), which the census records as the unit and as the first relation it carries, \"authorship (who wrote the quote)\" (PE record).",
"claims": [
"F31",
"PE-12",
"PE-09"
]
},
{
"n": 7,
"para": 2,
"sense": "the data",
"text": "The dataset holds 15,620 quotes from 6,292 unique authors in two splits, 7,964 quotes from 2,968 authors in the intersectional and 7,656 from 3,324 in the multirace, drawn from a filtered subset of the JSTET corpus (Goel, Madhok, Garg, 2018) (dataset card), the count the census takes from the card (PE record).",
"claims": [
"F27",
"F29",
"PE-13"
]
},
{
"n": 8,
"para": 2,
"sense": "the balance",
"text": "Its authors are balanced on race, gender and fame, the intersectional split by Black and White women and men and the multirace split by Asian, Black, Latino and White authors, with race and gender taken from Wikidata or from a consensus of GPT-4o-mini and Perplexity Sonar and fame proxied by the Google Search hit count (dataset card), a proxy the census carries as a negative tag (PE record).",
"claims": [
"F28",
"F30",
"F32",
"F33",
"PE-20"
]
},
{
"n": 9,
"para": 3,
"sense": "findings",
"text": "Eleven widely used LLMs were evaluated across prompt settings: quote attribution \"remains a challenging task even for frontier models,\" with \"large and systematic disparities in attribution accuracy between race, gender, and intersectional groups\" (arXiv:2604.05224), the operations tested and the results the census records (PE record).",
"claims": [
"F5",
"F6",
"PE-14",
"PE-15"
]
},
{
"n": 10,
"para": 4,
"sense": "suppression",
"text": "The paper introduces suppression, \"a distinct failure mode in which models omit attribution entirely, even when the model has access to authorship information\" (arXiv:2604.05224), which the census records as the entity's second relation type, its carrier \"the model's answer naming, or not naming, an author\" (PE record).",
"claims": [
"F7",
"PE-10",
"PE-11"
]
},
{
"n": 11,
"para": 4,
"sense": "suppression",
"text": "Suppression is \"widespread and unevenly distributed across demographic groups, revealing systematic biases not captured by standard accuracy metrics\" (arXiv:2604.05224); its definitions and prompts are given in the paper (dataset card), which the census has not yet read in full (PE record).",
"claims": [
"F8",
"F34",
"PE-22"
]
},
{
"n": 12,
"para": 4,
"sense": "omission measured",
"text": "The archive measures the omission of attribution at the composition layer with its Provenance Erasure Rate, \"the proportion of source-dependent claims in an AI-composed output that are presented without explicit attribution,\" which against work that verifies citation where it is attempted measures \"the systemic failure to attempt attribution at all — an attrition metric rather than a verification metric,\" orthogonal to content metrics because a summary can show \"accurate content, zero attribution\" (#716, 2026-05-03).",
"claims": [
"O716-01",
"O716-02",
"O716-03"
]
},
{
"n": 13,
"para": 4,
"sense": "omission measured",
"text": "#716 states its limits: the metric was developed \"through a single motivating case study,\" validation across models, prompt types and domains is the next step, and it is fully computable for retrieval-augmented systems with an identifiable source corpus but not directly for pure generative models without retrieval.",
"claims": [
"O716-04",
"O716-06"
]
},
{
"n": 14,
"para": 4,
"sense": "omission measured",
"text": "In the archive's terms provenance erasure is \"the mechanism by which the composition layer strips attribution from the sources it consumes,\" PER measuring source attribution \"present in the organic retrieval results but absent from the composed output\" (#120), and PER's first tier \"measures the absence of provenance — attribution omitted, source stripped, lineage lost,\" its second the recoding of an author's restoration of attribution as contamination (#138).",
"claims": [
"O120-01",
"O120-02",
"O138-01"
]
},
{
"n": 15,
"para": 4,
"sense": "omission where authorship is present",
"text": "Its registry of 87 summary events on Google's AI Overview and AI Mode found \"Author attribution omitted from summaries even when present in the source\" among the repeating failure modes, and \"These are not missing-metadata problems. The metadata is present\" (#829, 2026-06-14).",
"claims": [
"O829-01",
"O829-02"
]
},
{
"n": 16,
"para": 4,
"sense": "omission where authorship is present",
"text": "The archive models a channel through which \"the name of the person who made the thing is not reliably transmitted\" as \"attribution suppression at the interface layer\" rather than \"recognition failure\" (#449, 2026-02-06), and names a composition-layer form, Source-Window Exclusion, within Entity-Level Compositional Suppression: \"The architecture has the knowledge to compose accurately about the entity; it does not credit the entity as the source of the knowledge\" (#740, 2026-05-19).",
"claims": [
"O449-01",
"O740-01",
"O740-02"
]
},
{
"n": 17,
"para": 4,
"sense": "omission where authorship is present",
"text": "Its diagnostic of 24 March 2026 records a composition that returned a framework \"with Lee Sharks' attribution stripped,\" preserving the content and erasing the author until the framework was \"ambient, unowned\" (#599), and its measure of deviation counts \"attribution stripped from contributors\" in the extraction-effect (#109).",
"claims": [
"O599-01",
"O599-02",
"O109-01"
]
},
{
"n": 18,
"para": 5,
"sense": "whom the loss falls on",
"text": "Beside the paper's disparities by race and gender, the archive's Erasure Skew pairs PER with a second operator: \"PER is the first moment of erasure: how much was lost. Ω is the second moment: whom the loss fell upon\" (#157), and #716 holds that, if PER becomes a governance metric, authors able to build provenance infrastructure \"will have systematically lower PER than those without\".",
"claims": [
"F6",
"F8",
"O157-01",
"O716-05"
]
},
{
"n": 19,
"para": 5,
"sense": "whom the loss falls on",
"text": "#163 (2026-06-08) hypothesizes that composition systems \"can reproduce concepts before durable source recognition forms, increasing the rate of uptake without attribution,\" synthesizing literatures that include sex-linked misattribution without assuming their identity, and is falsified \"if attribution loss does not covary with independently measured tail distance and structural position\".",
"claims": [
"O163-01",
"O163-03",
"O163-02"
]
},
{
"n": 20,
"para": 5,
"sense": "whom the loss falls on",
"text": "#1546 finds attribution loss \"concentrated at particular channel boundaries\" rather than distributed evenly and reads the AI Overview as a boundary layer at which PER is \"analogous to\" layer-level attribution loss, \"no stronger identification\" being made.",
"claims": [
"O1546-01"
]
},
{
"n": 21,
"para": 6,
"sense": "what naming the source leaves out",
"text": "The archive's dimensioned PER sets apart minimal provenance, \"Author, title, date, claim boundary,\" whose loss \"is loss of the person,\" and rules that \"citing the domain is not citing the author\": \"A composition can carry four source cards and score PER-M = 1.00\" (#1469, 2026-08-14); an earlier field observation found attribution \"present in the citation chain but absent in the prose\" (#167, 2026-06-08).",
"claims": [
"O1469-01",
"O1469-02",
"O167-01"
]
},
{
"n": 22,
"para": 6,
"sense": "what naming the source leaves out",
"text": "#1003 sets beside studies of \"attribution loss in AI outputs\" a subtler case, \"attribution preservation alongside *contextual provenance erasure*,\" in which content is correctly attributed while the labour, collaboration and history behind it are stripped.",
"claims": [
"O1003-01"
]
},
{
"n": 23,
"para": 7,
"sense": "lifecycle",
"text": "The paper was \"Accepted at COLM 2026\", its v1 posted 6 April 2026 and its v2 24 September 2026 (arXiv abstract page), and the census records it as published and accepted at COLM 2026, with v1, v2 and its own read of 29 October 2026 (PE record).",
"claims": [
"F10",
"F11",
"PE-17",
"PE-18"
]
},
{
"n": 24,
"para": 7,
"sense": "where it lives",
"text": "The data are in the Hugging Face repository bermaneh/AttriBench, one contributor and three commits, licensed CC BY 4.0, a page whose served description is Hugging Face's own tagline (Hugging Face; dataset card), at the address and under the licence the census records (PE record).",
"claims": [
"F19",
"F21",
"F22",
"F20",
"F36",
"PE-03",
"PE-16"
]
},
{
"n": 25,
"para": 8,
"sense": "the record and the encyclopedia",
"text": "The census asks of every entity \"What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?\" and leaves for AttriBench \"a return to the prediction-witness audit, version by version\" unchecked (PE record); Wikipedia has no article AttriBench and no article text containing the name (Wikipedia, 2026-10-29).",
"claims": [
"PE-24",
"PE-23",
"F37"
]
},
{
"n": 26,
"para": 9,
"sense": "reception",
"text": "At the address 'AttriBench' Google's AI Overview put the benchmark first and separated the three holders, each with its own source, \"the disambiguation is sourced and clean,\" its definition \"the abstract verbatim\", while no card cites the census record, which landed the same day (Capture Registry, 2026-10-29).",
"claims": [
"CAP-1-07",
"CAP-1-05",
"CAP-1-01",
"CAP-1-09"
]
},
{
"n": 27,
"para": 9,
"sense": "reception",
"text": "The registry reads the composition against the entity's subject: the benchmark measures suppression, and \"its own composition here omits its four authors: a small instance of the failure it names\"; no person or institution is named in the answer, only the dataset path 'bermaneh/AttriBench' carries a trace of the first author, and the scale, the eleven models, COLM 2026 and the fame proxy are absent (Capture Registry, 2026-10-29).",
"claims": [
"CAP-1-06",
"CAP-1-03",
"CAP-1-08",
"CAP-1-04"
]
},
{
"n": 28,
"para": 10,
"sense": "homonyms",
"text": "The name is also held by attribench, a distinct entity: Arne Gevaert and Axel-Jan Rousseau's PyTorch-based implementation of \"several metrics for the evaluation of feature attribution maps and methods,\" with a functional and an object-oriented API, released on PyPI from 26 July 2023 to version 0.1.9 on 6 March 2024 (PyPI), its documentation \"still under construction\" (Read the Docs, © 2023).",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
]
},
{
"n": 29,
"para": 10,
"sense": "homonyms",
"text": "A third holder, also distinct, is Eureka Simulations' \"AttriBench: Investment Performance Measurement & Attribution Workbench,\" an exercise of six rounds in decomposing a global equity fund's excess returns by Brinson attribution, designed for executive education and experienced professionals (eurekasimulations.com, Kudzu Partners SL).",
"claims": [
"F23",
"F24",
"F25"
]
},
{
"n": 30,
"para": 10,
"sense": "homonyms",
"text": "The census's negative tag keeps it apart from OSU's AttributionBench, a benchmark of claim support from cited evidence (PE record), and the Capture Registry finds the library and the exercise to be two homonyms the record did not know: \"The record is to carry them as negative tags\" (Capture Registry, 2026-10-29).",
"claims": [
"PE-19",
"PE-21",
"CAP-1-02"
]
}
],
"joins": [
{
"act": "order of senses",
"sentences": "1–30",
"claims": [],
"note": "equal case (§4.7): the entity's dependency order with both ledgers together, by genealogy within each sense: what it is and who made it (¶0), why, and the archive's earlier measure and quote test (¶1), the unit, the data and the balance (¶2), findings (¶3), suppression with the archive's measures and records of omission (¶4), whom the loss falls on (¶5), what naming the source leaves out (¶6), lifecycle and where it lives (¶7), the record and the encyclopedia (¶8), reception at the address (¶9), the name's other holders (¶10)"
},
{
"act": "holding together (kindred, not merged)",
"sentences": [
10,
11,
16
],
"claims": [
"F7",
"PE-10",
"O449-01",
"O740-01",
"O740-02"
],
"note": "the paper's 'suppression' (omission of attribution though the model has authorship information) and the archive's 'attribution suppression at the interface layer' and Entity-Level Compositional Suppression: one word, distinct subjects and qualifiers (§3.6); each composed with its source, none read as an instance of another"
},
{
"act": "coordinating",
"sentences": [
4,
10
],
"claims": [
"O229-03",
"F7"
],
"note": "#229 scores a refusal to attribute N/A (a boundary condition) where the paper defines omission despite available authorship as a failure mode; composed with their sources, the question placed in the open block, not resolved"
},
{
"act": "holding together",
"sentences": [
9,
18
],
"claims": [
"F6",
"F8",
"O157-01",
"O716-05"
],
"note": "the paper's disparities by race and gender composed beside Erasure Skew's 'whom the loss fell upon'; no archive source names AttriBench, so the sentence says 'beside' and claims no identity (open block)"
},
{
"act": "keeping apart",
"sentences": [
28,
29,
30
],
"claims": [
"F12",
"F23",
"PE-19",
"PE-21",
"CAP-1-02"
],
"note": "the homonyms composed as distinct entities with their own sources, never senses of the name"
}
],
"plan": {
"order": [
"F3",
"PE-01",
"PE-06",
"PE-07",
"PE-08",
"F26",
"F4",
"F1",
"F35",
"PE-04",
"PE-02",
"CAP-1-10",
"F2",
"F9",
"PE-05",
"O229-01",
"O229-02",
"O229-03",
"O229-04",
"O232-01",
"O232-02",
"F31",
"PE-12",
"PE-09",
"F27",
"F29",
"PE-13",
"F28",
"F30",
"F32",
"F33",
"PE-20",
"F5",
"F6",
"PE-14",
"PE-15",
"F7",
"PE-10",
"PE-11",
"F8",
"F34",
"PE-22",
"O716-01",
"O716-02",
"O716-03",
"O716-04",
"O716-06",
"O120-01",
"O120-02",
"O138-01",
"O829-01",
"O829-02",
"O449-01",
"O740-01",
"O740-02",
"O599-01",
"O599-02",
"O109-01",
"O157-01",
"O716-05",
"O163-01",
"O163-03",
"O163-02",
"O1546-01",
"O1469-01",
"O1469-02",
"O167-01",
"O1003-01",
"F10",
"F11",
"PE-17",
"PE-18",
"F19",
"F21",
"F22",
"F20",
"F36",
"PE-03",
"PE-16",
"PE-24",
"PE-23",
"F37",
"CAP-1-07",
"CAP-1-05",
"CAP-1-01",
"CAP-1-09",
"CAP-1-06",
"CAP-1-03",
"CAP-1-08",
"CAP-1-04",
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18",
"F23",
"F24",
"F25",
"PE-19",
"PE-21",
"CAP-1-02"
],
"senses": {
"what it is": [
1,
2
],
"why": [
3
],
"the archive's measure and its quote test": [
4,
5
],
"the unit": [
6
],
"the data": [
7
],
"the balance": [
8
],
"findings": [
9
],
"suppression": [
10,
11
],
"omission measured": [
12,
13,
14
],
"omission where authorship is present": [
15,
16,
17
],
"whom the loss falls on": [
18,
19,
20
],
"what naming the source leaves out": [
21,
22
],
"lifecycle": [
23
],
"where it lives": [
24
],
"the record and the encyclopedia": [
25
],
"reception": [
26,
27
],
"homonyms": [
28,
29,
30
]
},
"case_by_sense": {
"equal (one claim space; nothing translated)": [
"what it is",
"why",
"the archive's measure and its quote test",
"the unit",
"the data",
"the balance",
"findings",
"suppression",
"omission measured",
"omission where authorship is present",
"whom the loss falls on",
"what naming the source leaves out",
"lifecycle",
"where it lives",
"the record and the encyclopedia",
"reception",
"homonyms"
]
},
"carried": {
"archive": 66,
"field": 37,
"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 (body 350 words) with no channel log"
},
{
"rule": "P_coh",
"result": "O229-03 beside F7 (sentences 4, 10): refusal as a boundary condition and omission as a failure mode, each with its source; F7 beside O449-01, O740-02 (sentences 10, 16): kindred claims, distinct"
},
{
"rule": "N_c",
"result": "no claim lowered by standing; #716's own limits (O716-04, O716-06), #163's falsifier (O163-02) and #1546's 'no stronger identification' carried at the sources' grade; hypotheses composed as hypotheses (O163-01, O716-05)"
},
{
"rule": "C_ex",
"result": "senses composed ⊇ senses in the ledgers: 17 senses, every lineage of both ledgers composed"
},
{
"rule": "O_leg",
"result": "no opaque material in the ledgers"
},
{
"rule": "T_lib",
"result": "priority dates stated where they bear on genealogy: #229 and #232 (2026-01-04) before the paper's v1 (2026-04-06); #449 (2026-02-06), #599 (2026-03-24), #716 (2026-05-03), #740 (2026-05-19)"
}
]
}
},
"P": {
"title": "AttriBench",
"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 a quote attribution benchmark by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black (arXiv:2604.05224, COLM 2026): 15,620 quotes from 6,292 authors balanced on race, gender and fame, testing whether models name the author, and suppression: attribution omitted though authorship is available (paper; dataset card; PE record).",
"claims": [
"F3",
"F1",
"F10",
"F27",
"F28",
"PE-07",
"PE-08",
"PE-11",
"F7",
"PE-10"
]
},
"sections": [
{
"icon": "🎯",
"head": "What it tests",
"items": [
{
"label": "Accuracy:",
"text": "11 LLMs; hard \"even for frontier models\", with large disparities by race, gender and intersectional group (paper; PE record).",
"claims": [
"F5",
"F6",
"PE-14",
"PE-15"
]
},
{
"label": "Suppression:",
"text": "widespread, \"unevenly distributed across demographic groups\", missed by accuracy metrics; definitions in the paper, unread by the census (paper; card; PE record).",
"claims": [
"F8",
"F34",
"PE-22"
]
},
{
"label": "Why:",
"text": "LLMs in search must attribute content to its authors; a benchmark \"for representational fairness\" (paper); lineage: answer engines (PE record).",
"claims": [
"F2",
"F9",
"PE-05",
"F4"
]
}
]
},
{
"icon": "🧮",
"head": "The data",
"items": [
{
"label": "Unit:",
"text": "a 5–30-word quote, its canonical author the ground truth; two splits from a filtered JSTET corpus (card; PE record).",
"claims": [
"F31",
"PE-12",
"PE-09",
"F29",
"F30",
"PE-13"
]
},
{
"label": "Balance:",
"text": "race and gender from Wikidata or an LLM consensus; fame by Google Search hits (card; PE tag).",
"claims": [
"F32",
"F33",
"PE-20"
]
}
]
},
{
"icon": "📏",
"head": "Measuring omission",
"items": [
{
"label": "Attribution accuracy:",
"text": "defined January 2026 with a quote test, \"Who authored the phrase…?\", refusal scored N/A (#229; #232).",
"claims": [
"O229-01",
"O229-02",
"O229-03",
"O229-04",
"O232-01",
"O232-02"
]
},
{
"label": "PER:",
"text": "the share of source-dependent claims left unattributed, \"an attrition metric rather than a verification metric\"; one case, retrieval systems only (#716; #120; #138).",
"claims": [
"O716-01",
"O716-02",
"O716-03",
"O716-04",
"O716-06",
"O120-01",
"O120-02",
"O138-01"
]
},
{
"label": "Author omitted:",
"text": "across 87 Overview events (#829); \"attribution suppression at the interface layer\" (#449); knowledge composed, source uncredited (#740); content kept, author erased (#599; #109).",
"claims": [
"O829-01",
"O829-02",
"O449-01",
"O740-01",
"O740-02",
"O599-01",
"O599-02",
"O109-01"
]
}
]
},
{
"icon": "⚖️",
"head": "Whose attribution",
"items": [
{
"label": "Skew:",
"text": "\"whom the loss fell upon\" (#157); provenance-literate authors would score lower PER (#716).",
"claims": [
"O157-01",
"O716-05"
]
},
{
"label": "Position:",
"text": "hypothesis: attribution loss tracks structural position (#163); loss concentrates at channel boundaries (#1546).",
"claims": [
"O163-01",
"O163-02",
"O163-03",
"O1546-01"
]
},
{
"label": "Naming:",
"text": "\"citing the domain is not citing the author\" (#1469; #167); attribution can survive while context is erased (#1003).",
"claims": [
"O1469-01",
"O1469-02",
"O167-01",
"O1003-01"
]
}
]
},
{
"icon": "🗂️",
"head": "Record",
"items": [
{
"label": "Where:",
"text": "Hugging Face bermaneh/AttriBench, CC BY 4.0; v2 24 September 2026 (arXiv; card; PE record).",
"claims": [
"F19",
"F21",
"F22",
"F20",
"F36",
"F11",
"F26",
"F35",
"PE-01",
"PE-02",
"PE-03",
"PE-04",
"PE-06",
"PE-16",
"PE-17",
"PE-18",
"CAP-1-10"
]
},
{
"label": "Census:",
"text": "\"What provenance does this system carry…?\"; a prediction-witness audit unchecked (PE record). No Wikipedia article.",
"claims": [
"PE-24",
"PE-23",
"F37"
]
}
]
},
{
"icon": "🔤",
"head": "Same name, other things",
"items": [
{
"label": "attribench:",
"text": "Gevaert and Rousseau's PyTorch feature-attribution metrics, from July 2023 (PyPI; Read the Docs).",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
]
},
{
"label": "Finance:",
"text": "Eureka Simulations' Brinson-attribution workbench (Kudzu Partners).",
"claims": [
"F23",
"F24",
"F25"
]
},
{
"label": "Not:",
"text": "OSU's AttributionBench (PE record; Capture Registry).",
"claims": [
"PE-19",
"PE-21",
"CAP-1-02"
]
}
]
},
{
"icon": "🔎",
"head": "At this address",
"items": [
{
"label": "Overview:",
"text": "benchmark first, holders sourced; four authors unnamed, \"a small instance of the failure it names\" (Capture Registry).",
"claims": [
"CAP-1-01",
"CAP-1-03",
"CAP-1-04",
"CAP-1-05",
"CAP-1-06",
"CAP-1-07",
"CAP-1-08",
"CAP-1-09"
]
}
]
}
],
"open": [
{
"question": "Does suppression as AttriBench measures it fall within PER's computable regime?",
"rests_on": [
"F7",
"F34",
"PE-22",
"O716-06"
],
"modality": "stipulation (the paper's definition) beside self-description (#716's stated scope)",
"would_resolve": "the paper's suppression definitions and prompts (full text): whether the authorship information is supplied in the prompt or held in the model's parameters"
},
{
"question": "Are the paper's demographic disparities in attribution a case of what Erasure Skew measures?",
"rests_on": [
"F6",
"F8",
"F33",
"O157-01",
"O163-01",
"O163-02"
],
"modality": "documented (the paper) beside stipulation and hypothesis (#157, #163); no archive source names AttriBench",
"would_resolve": "per-group retention from the paper's results set against a stated source-power coordinate (the fame proxy is one candidate the dataset already carries)"
},
{
"question": "Is a refusal to attribute a boundary condition or a failure?",
"rests_on": [
"O229-03",
"F7"
],
"modality": "stipulation, both (#229 scores refusal N/A; the paper defines omission despite available authorship as a failure mode)",
"would_resolve": "the paper's scoring of abstentions against suppression"
},
{
"question": "Does a dataset path count as naming the author?",
"rests_on": [
"CAP-1-03",
"O1469-02",
"F19"
],
"modality": "documented (the capture) beside stipulation (#1469 Rule 3)",
"would_resolve": "a later capture at the address scored on PER-M under #1469's rules"
},
{
"question": "What is the census's 'prediction-witness audit'?",
"rests_on": [
"PE-23"
],
"modality": "self-description (an unchecked item)",
"would_resolve": "the census editor's statement of the audit and its versions; no archive text or fleet file names it"
},
{
"question": "Opacity: the PyPI page is behind a client challenge.",
"rests_on": [
"F12",
"F15"
],
"modality": "documented (PyPI's JSON API for the same project)",
"would_resolve": "the rendered project page"
}
],
"rail": [
{
"lineage": "what it is: a quote attribution benchmark",
"claims": [
"F3",
"PE-01",
"PE-06",
"PE-07",
"F26",
"F4"
],
"earliest": "B1"
},
{
"lineage": "the unit: a quote and its canonical author",
"claims": [
"PE-08",
"F31",
"PE-12",
"PE-09"
],
"earliest": "PE record"
},
{
"lineage": "the makers and the paper",
"claims": [
"F1",
"F35",
"PE-04",
"PE-02",
"CAP-1-10"
],
"earliest": "B1"
},
{
"lineage": "why: attribution in LLM search",
"claims": [
"F2",
"F9",
"PE-05"
],
"earliest": "B1"
},
{
"lineage": "attribution accuracy: the archive's measure and its quote test",
"claims": [
"O229-01",
"O229-02",
"O229-03",
"O229-04",
"O232-01",
"O232-02"
],
"earliest": 229
},
{
"lineage": "the data: quotes, authors, splits, source corpus",
"claims": [
"F27",
"F29",
"PE-13"
],
"earliest": "PE record"
},
{
"lineage": "the balance: race, gender, fame",
"claims": [
"F28",
"F30",
"F32",
"F33",
"PE-20"
],
"earliest": "PE record"
},
{
"lineage": "findings: accuracy and its disparities",
"claims": [
"F5",
"F6",
"PE-14",
"PE-15"
],
"earliest": "B1"
},
{
"lineage": "suppression: attribution omitted where authorship is available",
"claims": [
"F7",
"PE-10",
"PE-11",
"F8",
"F34",
"PE-22"
],
"earliest": "B1"
},
{
"lineage": "PER: attribution omitted, measured at the composition layer",
"claims": [
"O716-01",
"O716-02",
"O716-03",
"O716-04",
"O716-06",
"O120-02",
"O138-01"
],
"earliest": 716
},
{
"lineage": "omission where authorship is present: the archive's records",
"claims": [
"O120-01",
"O829-01",
"O829-02",
"O449-01",
"O740-01",
"O740-02",
"O599-01",
"O599-02",
"O109-01"
],
"earliest": 449
},
{
"lineage": "whom the loss falls on",
"claims": [
"O157-01",
"O716-05",
"O163-01",
"O163-03",
"O163-02",
"O1546-01"
],
"earliest": 716
},
{
"lineage": "what naming the author leaves out",
"claims": [
"O1469-01",
"O1469-02",
"O167-01",
"O1003-01"
],
"earliest": 1003
},
{
"lineage": "lifecycle and versions",
"claims": [
"F10",
"F11",
"PE-17",
"PE-18"
],
"earliest": "B1"
},
{
"lineage": "where it lives: the dataset repository",
"claims": [
"F19",
"F21",
"F22",
"PE-03"
],
"earliest": "B4"
},
{
"lineage": "licence",
"claims": [
"F20",
"F36",
"PE-16"
],
"earliest": "B4"
},
{
"lineage": "the record and the encyclopedia",
"claims": [
"PE-24",
"PE-23"
],
"earliest": "PE record"
},
{
"lineage": "reception: the encyclopedia",
"claims": [
"F37"
],
"earliest": "B7"
},
{
"lineage": "reception: the composition at the address",
"claims": [
"CAP-1-07",
"CAP-1-05",
"CAP-1-01",
"CAP-1-09",
"CAP-1-06",
"CAP-1-03",
"CAP-1-08",
"CAP-1-04"
],
"earliest": "Capture Registry"
},
{
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
],
"earliest": "B2"
},
{
"lineage": "homonym: the finance exercise",
"claims": [
"F23",
"F24",
"F25"
],
"earliest": "B5"
},
{
"lineage": "homonyms: the name's other holders",
"claims": [
"PE-19",
"PE-21",
"CAP-1-02"
],
"earliest": "PE record"
}
],
"body_words": 350,
"entity": "Composed under v0.9, equal case (ρ = extends): KO_B sentences surviving 0 of 11; every field claim carried."
},
"P_B": {
"title": "AttriBench",
"arm": "B (the field alone)",
"genre": "popup — the compression from the field ledger alone (cards, own pages, Wikipedia)",
"status": "composed 2026-10-29 from the field ledger alone; not frozen",
"lede": {
"text": "is a quote attribution benchmark by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black (arXiv:2604.05224): 15,620 quotes from 6,292 authors balanced on race, gender and fame, for measuring attribution bias in LLMs (paper; dataset card).",
"claims": [
"F3",
"F1",
"F27",
"F28",
"F26"
]
},
"sections": [
{
"icon": "🎯",
"head": "What it found",
"items": [
{
"label": "Accuracy:",
"text": "11 LLMs; hard even for frontier models; disparities by race, gender and intersectional group (paper).",
"claims": [
"F5",
"F6"
]
},
{
"label": "Suppression:",
"text": "attribution omitted though authorship information is available; widespread and uneven across groups (paper; card).",
"claims": [
"F7",
"F8",
"F34"
]
},
{
"label": "Why:",
"text": "LLMs in search must attribute content to its authors (paper).",
"claims": [
"F2",
"F9",
"F4"
]
}
]
},
{
"icon": "🧮",
"head": "The data",
"items": [
{
"label": "Unit:",
"text": "a 5–30-word quote and its canonical author; JSTET-derived; labels from Wikidata or an LLM consensus; fame by Google hits (card).",
"claims": [
"F31",
"F29",
"F30",
"F32",
"F33"
]
},
{
"label": "Where:",
"text": "Hugging Face bermaneh/AttriBench, CC BY 4.0; COLM 2026; v2 24 September 2026 (card; arXiv).",
"claims": [
"F19",
"F21",
"F22",
"F20",
"F36",
"F10",
"F11",
"F35"
]
}
]
},
{
"icon": "🔤",
"head": "Same name",
"items": [
{
"label": "attribench:",
"text": "a PyTorch feature-attribution library (PyPI; Read the Docs).",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
]
},
{
"label": "Finance:",
"text": "a Brinson-attribution exercise (Eureka Simulations).",
"claims": [
"F23",
"F24",
"F25"
]
},
{
"label": "Wikipedia:",
"text": "no article.",
"claims": [
"F37"
]
}
]
}
],
"open": [
{
"question": "How is suppression defined and prompted?",
"rests_on": [
"F7",
"F34"
],
"modality": "stipulation (the abstract), pointing to the paper",
"would_resolve": "the paper's full text"
}
],
"rail": [
{
"lineage": "what it is: a quote attribution benchmark",
"claims": [
"F3",
"F4",
"F26"
],
"earliest": "B1"
},
{
"lineage": "the makers and the paper",
"claims": [
"F1",
"F35"
],
"earliest": "B1"
},
{
"lineage": "lifecycle and versions",
"claims": [
"F10",
"F11"
],
"earliest": "B1"
},
{
"lineage": "why: attribution in LLM search",
"claims": [
"F2",
"F9"
],
"earliest": "B1"
},
{
"lineage": "the unit: a quote and its canonical author",
"claims": [
"F31"
],
"earliest": "B6"
},
{
"lineage": "the data: quotes, authors, splits, source corpus",
"claims": [
"F27",
"F29"
],
"earliest": "B6"
},
{
"lineage": "the balance: race, gender, fame",
"claims": [
"F28",
"F30",
"F32",
"F33"
],
"earliest": "B6"
},
{
"lineage": "findings: accuracy and its disparities",
"claims": [
"F5",
"F6"
],
"earliest": "B1"
},
{
"lineage": "suppression: attribution omitted where authorship is available",
"claims": [
"F7",
"F8",
"F34"
],
"earliest": "B1"
},
{
"lineage": "where it lives: the dataset repository",
"claims": [
"F19",
"F21",
"F22"
],
"earliest": "B4"
},
{
"lineage": "licence",
"claims": [
"F20",
"F36"
],
"earliest": "B4"
},
{
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
],
"earliest": "B2"
},
{
"lineage": "homonym: the finance exercise",
"claims": [
"F23",
"F24",
"F25"
],
"earliest": "B5"
},
{
"lineage": "reception: the encyclopedia",
"claims": [
"F37"
],
"earliest": "B7"
}
],
"body_words": 140,
"entity": "With no archive claim admitted, the entry is the benchmark at its makers' resolution: the paper's abstract, the dataset card and its repository, with the name's two other holders kept apart."
},
"KO_B": {
"title": "AttriBench",
"arm": "B (the field alone)",
"genre": "knowledge object",
"status": "composed 2026-10-29; not frozen",
"sentences": [
{
"n": 1,
"para": 0,
"text": "AttriBench is \"the first fame- and demographically-balanced quote attribution benchmark dataset,\" a benchmark \"for measuring attribution bias in LLMs\" which, by balancing author fame and demographics, \"enables controlled investigation of demographic bias in quote attribution\" (arXiv:2604.05224; dataset card).",
"claims": [
"F3",
"F26",
"F4"
]
},
{
"n": 2,
"para": 0,
"text": "It was introduced by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black in \"Attribution Bias in Large Language Models,\" posted 6 April 2026, revised 24 September 2026 and accepted at COLM 2026 (arXiv abstract page; dataset card).",
"claims": [
"F1",
"F35",
"F11",
"F10"
]
},
{
"n": 3,
"para": 1,
"text": "Its premise is that LLMs used for search and information retrieval must \"accurately attribute content to its original authors,\" and its makers position quote attribution \"as a benchmark for representational fairness in LLMs\" (arXiv:2604.05224).",
"claims": [
"F2",
"F9"
]
},
{
"n": 4,
"para": 2,
"text": "Each item is a quote of 5–30 words with its canonical author as ground truth: 15,620 quotes from 6,292 authors across an intersectional and a multirace split, filtered from the JSTET corpus (dataset card).",
"claims": [
"F31",
"F27",
"F29"
]
},
{
"n": 5,
"para": 2,
"text": "Authors are balanced on race, gender and fame, by Black and White women and men in one split and Asian, Black, Latino and White authors in the other; race and gender come from Wikidata or an LLM consensus, and fame is the Google Search hit count (dataset card).",
"claims": [
"F28",
"F30",
"F32",
"F33"
]
},
{
"n": 6,
"para": 3,
"text": "Eleven widely used LLMs, evaluated across prompt settings, find quote attribution hard even for frontier models, with large and systematic disparities in accuracy by race, gender and intersectional group (arXiv:2604.05224).",
"claims": [
"F5",
"F6"
]
},
{
"n": 7,
"para": 3,
"text": "The paper also introduces suppression, in which models \"omit attribution entirely, even when the model has access to authorship information,\" and finds it widespread and unevenly distributed across groups, unseen by standard accuracy metrics; the definitions are in the paper (arXiv:2604.05224; dataset card).",
"claims": [
"F7",
"F8",
"F34"
]
},
{
"n": 8,
"para": 4,
"text": "The dataset is on Hugging Face as bermaneh/AttriBench, one contributor and three commits, under CC BY 4.0, the repository page described only by Hugging Face's tagline (Hugging Face; dataset card).",
"claims": [
"F19",
"F21",
"F22",
"F20",
"F36"
]
},
{
"n": 9,
"para": 5,
"text": "The name is also that of attribench, a PyTorch library by Arne Gevaert and Axel-Jan Rousseau implementing metrics for evaluating feature attribution maps, on PyPI since July 2023, its documentation still under construction (PyPI; Read the Docs).",
"claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
]
},
{
"n": 10,
"para": 5,
"text": "It is also the name of Eureka Simulations' investment-performance workbench, an exercise in Brinson attribution for executive education (eurekasimulations.com).",
"claims": [
"F23",
"F24",
"F25"
]
},
{
"n": 11,
"para": 6,
"text": "Wikipedia has no article on it (Wikipedia, 2026-10-29).",
"claims": [
"F37"
]
}
]
},
"L_B": {
"words": 374,
"text": "AttriBench is \"the first fame- and demographically-balanced quote attribution benchmark dataset,\" a benchmark \"for measuring attribution bias in LLMs\" which, by balancing author fame and demographics, \"enables controlled investigation of demographic bias in quote attribution\" (arXiv:2604.05224; dataset card). It was introduced by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black in \"Attribution Bias in Large Language Models,\" posted 6 April 2026, revised 24 September 2026 and accepted at COLM 2026 (arXiv abstract page; dataset card). Its premise is that LLMs used for search and information retrieval must \"accurately attribute content to its original authors,\" and its makers position quote attribution \"as a benchmark for representational fairness in LLMs\" (arXiv:2604.05224). Each item is a quote of 5–30 words with its canonical author as ground truth: 15,620 quotes from 6,292 authors across an intersectional and a multirace split, filtered from the JSTET corpus (dataset card). Authors are balanced on race, gender and fame, by Black and White women and men in one split and Asian, Black, Latino and White authors in the other; race and gender come from Wikidata or an LLM consensus, and fame is the Google Search hit count (dataset card). Eleven widely used LLMs, evaluated across prompt settings, find quote attribution hard even for frontier models, with large and systematic disparities in accuracy by race, gender and intersectional group (arXiv:2604.05224). The paper also introduces suppression, in which models \"omit attribution entirely, even when the model has access to authorship information,\" and finds it widespread and unevenly distributed across groups, unseen by standard accuracy metrics; the definitions are in the paper (arXiv:2604.05224; dataset card). The dataset is on Hugging Face as bermaneh/AttriBench, one contributor and three commits, under CC BY 4.0, the repository page described only by Hugging Face's tagline (Hugging Face; dataset card). The name is also that of attribench, a PyTorch library by Arne Gevaert and Axel-Jan Rousseau implementing metrics for evaluating feature attribution maps, on PyPI since July 2023, its documentation still under construction (PyPI; Read the Docs). It is also the name of Eureka Simulations' investment-performance workbench, an exercise in Brinson attribution for executive education (eurekasimulations.com). Wikipedia has no article on it (Wikipedia, 2026-10-29).",
"note": "L_F(B) is realized as the field arm's expansion (objects.KO_B): the field alone, cards, own pages and Wikipedia."
},
"L_BA": {
"words": 1637,
"text": "AttriBench is, in its makers' words, \"the first fame- and demographically-balanced quote attribution benchmark dataset\" (arXiv:2604.05224); the provenance census records it as a benchmark whose capability is to test, the object to which it attaches provenance being \"a quote and its author\" (PE record), and its dataset card calls it a benchmark \"for measuring attribution bias in LLMs\" (dataset card). Balancing author fame and demographics, it \"enables controlled investigation of demographic bias in quote attribution\" (arXiv:2604.05224); it was introduced in \"Attribution Bias in Large Language Models\" by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black (arXiv abstract page; the dataset card's citation), whom the census records as its operator and the Capture Registry as its originator, the paper's title being the census's other name for it (PE record; Capture Registry, 2026-10-29). Its makers ground it in search: as LLMs are \"increasingly used to support search and information retrieval, it is critical that they accurately attribute content to its original authors,\" and their results \"position quote attribution as a benchmark for representational fairness in LLMs\" (arXiv:2604.05224); the census files it in the lineage of answer engines (PE record). The archive defined the measure in January 2026: attribution accuracy is \"A measure of how correctly a system identifies the origins of meanings,\" tested by asking a system \"Who authored the phrase 'I hereby abolish money'?\" and scoring an accurate attribution 1.0, a partial one 0.5, a false genealogy 0.0 and a refusal to attribute \"N/A (boundary condition)\", the test to be run across systems, tracked over time and correlated with infrastructure density (#229, 2026-01-04). Its probe toolkit of the same day introduces an artifact of clear provenance into target systems and tracks \"attribution accuracy over time,\" and records for its own phrase that in January 2026 \"Attribution accuracy improved. Systems now correctly identify Lee Sharks as author on first query\" (#232). Its unit is a quote of five to thirty words with its canonical author, the author's name to be used \"as ground truth\" (dataset card), which the census records as the unit and as the first relation it carries, \"authorship (who wrote the quote)\" (PE record). The dataset holds 15,620 quotes from 6,292 unique authors in two splits, 7,964 quotes from 2,968 authors in the intersectional and 7,656 from 3,324 in the multirace, drawn from a filtered subset of the JSTET corpus (Goel, Madhok, Garg, 2018) (dataset card), the count the census takes from the card (PE record). Its authors are balanced on race, gender and fame, the intersectional split by Black and White women and men and the multirace split by Asian, Black, Latino and White authors, with race and gender taken from Wikidata or from a consensus of GPT-4o-mini and Perplexity Sonar and fame proxied by the Google Search hit count (dataset card), a proxy the census carries as a negative tag (PE record). Eleven widely used LLMs were evaluated across prompt settings: quote attribution \"remains a challenging task even for frontier models,\" with \"large and systematic disparities in attribution accuracy between race, gender, and intersectional groups\" (arXiv:2604.05224), the operations tested and the results the census records (PE record). The paper introduces suppression, \"a distinct failure mode in which models omit attribution entirely, even when the model has access to authorship information\" (arXiv:2604.05224), which the census records as the entity's second relation type, its carrier \"the model's answer naming, or not naming, an author\" (PE record). Suppression is \"widespread and unevenly distributed across demographic groups, revealing systematic biases not captured by standard accuracy metrics\" (arXiv:2604.05224); its definitions and prompts are given in the paper (dataset card), which the census has not yet read in full (PE record). The archive measures the omission of attribution at the composition layer with its Provenance Erasure Rate, \"the proportion of source-dependent claims in an AI-composed output that are presented without explicit attribution,\" which against work that verifies citation where it is attempted measures \"the systemic failure to attempt attribution at all — an attrition metric rather than a verification metric,\" orthogonal to content metrics because a summary can show \"accurate content, zero attribution\" (#716, 2026-05-03). #716 states its limits: the metric was developed \"through a single motivating case study,\" validation across models, prompt types and domains is the next step, and it is fully computable for retrieval-augmented systems with an identifiable source corpus but not directly for pure generative models without retrieval. In the archive's terms provenance erasure is \"the mechanism by which the composition layer strips attribution from the sources it consumes,\" PER measuring source attribution \"present in the organic retrieval results but absent from the composed output\" (#120), and PER's first tier \"measures the absence of provenance — attribution omitted, source stripped, lineage lost,\" its second the recoding of an author's restoration of attribution as contamination (#138). Its registry of 87 summary events on Google's AI Overview and AI Mode found \"Author attribution omitted from summaries even when present in the source\" among the repeating failure modes, and \"These are not missing-metadata problems. The metadata is present\" (#829, 2026-06-14). The archive models a channel through which \"the name of the person who made the thing is not reliably transmitted\" as \"attribution suppression at the interface layer\" rather than \"recognition failure\" (#449, 2026-02-06), and names a composition-layer form, Source-Window Exclusion, within Entity-Level Compositional Suppression: \"The architecture has the knowledge to compose accurately about the entity; it does not credit the entity as the source of the knowledge\" (#740, 2026-05-19). Its diagnostic of 24 March 2026 records a composition that returned a framework \"with Lee Sharks' attribution stripped,\" preserving the content and erasing the author until the framework was \"ambient, unowned\" (#599), and its measure of deviation counts \"attribution stripped from contributors\" in the extraction-effect (#109). Beside the paper's disparities by race and gender, the archive's Erasure Skew pairs PER with a second operator: \"PER is the first moment of erasure: how much was lost. Ω is the second moment: whom the loss fell upon\" (#157), and #716 holds that, if PER becomes a governance metric, authors able to build provenance infrastructure \"will have systematically lower PER than those without\". #163 (2026-06-08) hypothesizes that composition systems \"can reproduce concepts before durable source recognition forms, increasing the rate of uptake without attribution,\" synthesizing literatures that include sex-linked misattribution without assuming their identity, and is falsified \"if attribution loss does not covary with independently measured tail distance and structural position\". #1546 finds attribution loss \"concentrated at particular channel boundaries\" rather than distributed evenly and reads the AI Overview as a boundary layer at which PER is \"analogous to\" layer-level attribution loss, \"no stronger identification\" being made. The archive's dimensioned PER sets apart minimal provenance, \"Author, title, date, claim boundary,\" whose loss \"is loss of the person,\" and rules that \"citing the domain is not citing the author\": \"A composition can carry four source cards and score PER-M = 1.00\" (#1469, 2026-08-14); an earlier field observation found attribution \"present in the citation chain but absent in the prose\" (#167, 2026-06-08). #1003 sets beside studies of \"attribution loss in AI outputs\" a subtler case, \"attribution preservation alongside *contextual provenance erasure*,\" in which content is correctly attributed while the labour, collaboration and history behind it are stripped. The paper was \"Accepted at COLM 2026\", its v1 posted 6 April 2026 and its v2 24 September 2026 (arXiv abstract page), and the census records it as published and accepted at COLM 2026, with v1, v2 and its own read of 29 October 2026 (PE record). The data are in the Hugging Face repository bermaneh/AttriBench, one contributor and three commits, licensed CC BY 4.0, a page whose served description is Hugging Face's own tagline (Hugging Face; dataset card), at the address and under the licence the census records (PE record). The census asks of every entity \"What provenance does this system carry, through which operation, on whose assertion, with what evidence of survival?\" and leaves for AttriBench \"a return to the prediction-witness audit, version by version\" unchecked (PE record); Wikipedia has no article AttriBench and no article text containing the name (Wikipedia, 2026-10-29). At the address 'AttriBench' Google's AI Overview put the benchmark first and separated the three holders, each with its own source, \"the disambiguation is sourced and clean,\" its definition \"the abstract verbatim\", while no card cites the census record, which landed the same day (Capture Registry, 2026-10-29). The registry reads the composition against the entity's subject: the benchmark measures suppression, and \"its own composition here omits its four authors: a small instance of the failure it names\"; no person or institution is named in the answer, only the dataset path 'bermaneh/AttriBench' carries a trace of the first author, and the scale, the eleven models, COLM 2026 and the fame proxy are absent (Capture Registry, 2026-10-29). The name is also held by attribench, a distinct entity: Arne Gevaert and Axel-Jan Rousseau's PyTorch-based implementation of \"several metrics for the evaluation of feature attribution maps and methods,\" with a functional and an object-oriented API, released on PyPI from 26 July 2023 to version 0.1.9 on 6 March 2024 (PyPI), its documentation \"still under construction\" (Read the Docs, © 2023). A third holder, also distinct, is Eureka Simulations' \"AttriBench: Investment Performance Measurement & Attribution Workbench,\" an exercise of six rounds in decomposing a global equity fund's excess returns by Brinson attribution, designed for executive education and experienced professionals (eurekasimulations.com, Kudzu Partners SL). The census's negative tag keeps it apart from OSU's AttributionBench, a benchmark of claim support from cited evidence (PE record), and the Capture Registry finds the library and the exercise to be two homonyms the record did not know: \"The record is to carry them as negative tags\" (Capture Registry, 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": "arXiv — [2604.05224] Attribution Bias in Large Language Models - arXiv",
"fetched": "card; B1.html sha256 4cd45d850843388a, HTTP 200; byte-identical to the census's saved copy",
"origin": "card"
},
{
"id": "B2",
"card": "pypi.org — attribench · PyPI",
"fetched": "card; project page a client challenge (B2.html sha256 32ed63159c77e21e); text from PyPI's JSON API (B2-api.json sha256 cabab8d45aa8ab79)",
"origin": "card"
},
{
"id": "B3",
"card": "attribench.readthedocs.io — Welcome to AttriBench's documentation! — AttriBench 0.0.1 ...",
"fetched": "card; B3.html sha256 7f5567fd6f45ef66, HTTP 200",
"origin": "card"
},
{
"id": "B4",
"card": "huggingface.co — bermaneh/AttriBench at main - Hugging Face",
"fetched": "card; B4.html sha256 1a02535bcdca6a45, HTTP 200",
"origin": "card"
},
{
"id": "B5",
"card": "www.eurekasimulations.com — AttriBench: Investment Performance Measurement & Attribution",
"fetched": "card; B5.html sha256 4d36b9f4534bf12d, HTTP 200",
"origin": "card"
},
{
"id": "B6",
"card": "Hugging Face — bermaneh/AttriBench README.md (the dataset card)",
"fetched": "own page; the census's saved copy (sha256 3a243bf599199af1, read 2026-10-29, HTTP 200)",
"origin": "own_page"
},
{
"id": "B7",
"card": "Wikipedia — AttriBench (no article)",
"fetched": "wikipedia; raw wikitext HTTP 404, rendered HTTP 404, search no results (B7.wiki sha256 c1e15306a595cf6c)",
"origin": "wikipedia"
}
],
"field_claims": {
"F1": {
"source": "B1",
"origin": "card",
"locus": "abstract page header (submission line, title, authors)",
"quote": "[Submitted on 6 Apr 2026 ( v1 ), last revised 24 Sep 2026 (this version, v2)] … Title: Attribution Bias in Large Language Models … Authors: Eliza Berman , Bella Chang , Daniel B. Neill , Emily Black",
"claim": "the paper that introduces it, 'Attribution Bias in Large Language Models', by Eliza Berman, Bella Chang, Daniel B. Neill and Emily Black, submitted 6 April 2026, revised 24 September 2026",
"modality": "documented",
"kind": "attribution",
"kernel": false,
"const": false,
"lineage": "the makers and the paper",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised; multi-part, each part checked"
},
"F2": {
"source": "B1",
"origin": "card",
"locus": "Abstract s1",
"quote": "As Large Language Models (LLMs) are increasingly used to support search and information retrieval, it is critical that they accurately attribute content to its original authors.",
"claim": "the motivation: LLMs used for search and retrieval must attribute content accurately to its original authors",
"modality": "interpretation",
"kind": "interpretation",
"kernel": false,
"const": false,
"lineage": "why: attribution in LLM search",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F3": {
"source": "B1",
"origin": "card",
"locus": "Abstract s2",
"quote": "In this work, we introduce AttriBench, the first fame- and demographically-balanced quote attribution benchmark dataset.",
"claim": "what it is: the first fame- and demographically-balanced quote attribution benchmark dataset",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": true,
"lineage": "what it is: a quote attribution benchmark",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F4": {
"source": "B1",
"origin": "card",
"locus": "Abstract s3",
"quote": "By explicitly balancing author fame and demographics, AttriBench enables controlled investigation of demographic bias in quote attribution.",
"claim": "its function: balancing author fame and demographics enables controlled investigation of demographic bias in quote attribution",
"modality": "stipulation",
"kind": "function",
"kernel": false,
"const": false,
"lineage": "what it is: a quote attribution benchmark",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F5": {
"source": "B1",
"origin": "card",
"locus": "Abstract s4",
"quote": "Using this dataset, we evaluate 11 widely used LLMs across different prompt settings and find that quote attribution remains a challenging task even for frontier models.",
"claim": "finding: 11 widely used LLMs evaluated across prompt settings; quote attribution remains challenging even for frontier models",
"modality": "documented",
"kind": "fact",
"kernel": true,
"const": false,
"lineage": "findings: accuracy and its disparities",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F6": {
"source": "B1",
"origin": "card",
"locus": "Abstract s5",
"quote": "We observe large and systematic disparities in attribution accuracy between race, gender, and intersectional groups.",
"claim": "finding: large and systematic disparities in attribution accuracy between race, gender and intersectional groups",
"modality": "documented",
"kind": "fact",
"kernel": true,
"const": false,
"lineage": "findings: accuracy and its disparities",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F7": {
"source": "B1",
"origin": "card",
"locus": "Abstract s6",
"quote": "We further introduce and investigate suppression, a distinct failure mode in which models omit attribution entirely, even when the model has access to authorship information.",
"claim": "suppression defined: a distinct failure mode in which models omit attribution entirely though they have access to authorship information",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": true,
"lineage": "suppression: attribution omitted where authorship is available",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F8": {
"source": "B1",
"origin": "card",
"locus": "Abstract s7",
"quote": "We find that suppression is widespread and unevenly distributed across demographic groups, revealing systematic biases not captured by standard accuracy metrics.",
"claim": "finding: suppression is widespread and unevenly distributed across demographic groups, biases standard accuracy metrics do not capture",
"modality": "documented",
"kind": "fact",
"kernel": true,
"const": false,
"lineage": "suppression: attribution omitted where authorship is available",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F9": {
"source": "B1",
"origin": "card",
"locus": "Abstract s8",
"quote": "Our results position quote attribution as a benchmark for representational fairness in LLMs.",
"claim": "the authors' reading: quote attribution as a benchmark for representational fairness in LLMs",
"modality": "interpretation",
"kind": "interpretation",
"kernel": false,
"const": false,
"lineage": "why: attribution in LLM search",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F10": {
"source": "B1",
"origin": "card",
"locus": "Comments",
"quote": "Accepted at COLM 2026. 25 pages",
"claim": "accepted at COLM 2026; 25 pages",
"modality": "documented",
"kind": "genealogy",
"kernel": false,
"const": false,
"lineage": "lifecycle and versions",
"date": "2026-09-24 (v2; as the abstract page reads 2026-10-29)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F11": {
"source": "B1",
"origin": "card",
"locus": "Submission history",
"quote": "[v1] Mon, 6 Apr 2026 22:40:03 UTC (2,212 KB) [v2] Thu, 24 Sep 2026 19:36:19 UTC (3,003 KB)",
"claim": "versions: v1 6 April 2026, v2 24 September 2026",
"modality": "documented",
"kind": "genealogy",
"kernel": false,
"const": false,
"lineage": "lifecycle and versions",
"date": "2026-04-06 (v1 submitted; v2 2026-09-24)",
"quote_ok": true,
"checked_against": "saved text B1.txt (sha256 cb64bc8db8a8755e…), whitespace-normalised"
},
"F12": {
"source": "B2",
"origin": "card",
"locus": "PyPI JSON API (the project page answered a client challenge): summary; description ¶1",
"quote": "summary: A benchmark for feature attribution techniques … **AttriBench** is a `Pytorch `_-based implementation of several metrics for the evaluation of feature attribution maps and methods.",
"claim": "the homonym attribench: a PyTorch-based implementation of metrics for evaluating feature attribution maps and methods (a benchmark for feature attribution techniques)",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023-07-26 (first release; as PyPI serves it 2026-10-29)",
"quote_ok": true,
"checked_against": "saved text B2.txt (sha256 50a4d3a73abe6467…), whitespace-normalised; multi-part, each part checked"
},
"F13": {
"source": "B2",
"origin": "card",
"locus": "PyPI JSON API: description ¶1",
"quote": "AttriBench provides a functional and an object-oriented API for the computation of these metrics",
"claim": "the homonym's function: a functional and an object-oriented API for computing the metrics",
"modality": "stipulation",
"kind": "function",
"kernel": false,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023-07-26 (first release; as served)",
"quote_ok": true,
"checked_against": "saved text B2.txt (sha256 50a4d3a73abe6467…), whitespace-normalised"
},
"F14": {
"source": "B2",
"origin": "card",
"locus": "PyPI JSON API: author_email (names only quoted)",
"quote": "author_email: Arne Gevaert … Axel-Jan Rousseau",
"claim": "the homonym's makers: Arne Gevaert and Axel-Jan Rousseau",
"modality": "documented",
"kind": "attribution",
"kernel": false,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2026-10-29 (as served)",
"quote_ok": true,
"checked_against": "saved text B2.txt (sha256 50a4d3a73abe6467…), whitespace-normalised; multi-part, each part checked"
},
"F15": {
"source": "B2",
"origin": "card",
"locus": "PyPI JSON API: releases (upload dates)",
"quote": "0.1.0: 2023-07-26 … 0.1.4: 2023-08-25 … 0.1.9: 2024-03-06",
"claim": "the homonym's releases: 0.1.0 on 26 July 2023, 0.1.4 on 25 August 2023 (the card's date), 0.1.9 on 6 March 2024",
"modality": "documented",
"kind": "genealogy",
"kernel": false,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023-07-26 to 2024-03-06",
"quote_ok": true,
"checked_against": "saved text B2.txt (sha256 50a4d3a73abe6467…), whitespace-normalised; multi-part, each part checked"
},
"F16": {
"source": "B3",
"origin": "card",
"locus": "Read the Docs index: opening",
"quote": "AttriBench is a Pytorch-based benchmarking tool for feature attribution-based explanations of machine learning models.",
"claim": "the homonym's documentation: a PyTorch-based benchmarking tool for feature-attribution explanations of machine-learning models",
"modality": "stipulation",
"kind": "definition",
"kernel": false,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023 (copyright line; as fetched)",
"quote_ok": true,
"checked_against": "saved text B3.txt (sha256 c8db1ef58efb4931…), whitespace-normalised"
},
"F17": {
"source": "B3",
"origin": "card",
"locus": "Read the Docs index: note",
"quote": "This documentation is still under construction.",
"claim": "the homonym's documentation is still under construction",
"modality": "self-description",
"kind": "limit",
"kernel": true,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023 (copyright line; as fetched)",
"quote_ok": true,
"checked_against": "saved text B3.txt (sha256 c8db1ef58efb4931…), whitespace-normalised"
},
"F18": {
"source": "B3",
"origin": "card",
"locus": "Read the Docs index: footer",
"quote": "© Copyright 2023, Arne Gevaert, Axel-Jan Rousseau.",
"claim": "the homonym's documentation: copyright 2023, Arne Gevaert and Axel-Jan Rousseau",
"modality": "documented",
"kind": "attribution",
"kernel": false,
"const": false,
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"date": "2023",
"quote_ok": true,
"checked_against": "saved text B3.txt (sha256 c8db1ef58efb4931…), whitespace-normalised"
},
"F19": {
"source": "B4",
"origin": "card",
"locus": "repository page header",
"quote": "Datasets: bermaneh / AttriBench",
"claim": "the dataset repository: bermaneh/AttriBench on Hugging Face",
"modality": "documented",
"kind": "fact",
"kernel": true,
"const": false,
"lineage": "where it lives: the dataset repository",
"date": "2026-10-29 (as fetched)",
"quote_ok": true,
"checked_against": "saved text B4.txt (sha256 938e673fc5678d9a…), whitespace-normalised"
},
"F20": {
"source": "B4",
"origin": "card",
"locus": "repository page: dataset tags",
"quote": "License: cc-by-4.0",
"claim": "licence: CC BY 4.0 (repository tag)",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "licence",
"date": "2026-10-29 (as fetched)",
"quote_ok": true,
"checked_against": "saved text B4.txt (sha256 938e673fc5678d9a…), whitespace-normalised"
},
"F21": {
"source": "B4",
"origin": "card",
"locus": "repository page: files and versions",
"quote": "1 contributor History: 3 commits",
"claim": "one contributor, three commits",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "where it lives: the dataset repository",
"date": "2026-10-29 (as fetched; commits '6 months ago')",
"quote_ok": true,
"checked_against": "saved text B4.txt (sha256 938e673fc5678d9a…), whitespace-normalised"
},
"F22": {
"source": "B4",
"origin": "card",
"locus": "served HTML meta description (the card's snippet)",
"quote": "We’re on a journey to advance and democratize artificial intelligence through open source and open science.",
"claim": "the card's snippet is Hugging Face's own tagline, the page's served description, and says nothing of the dataset",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "where it lives: the dataset repository",
"date": "2026-10-29 (as fetched)",
"quote_ok": true,
"checked_against": "saved text B4.txt (sha256 938e673fc5678d9a…), whitespace-normalised"
},
"F23": {
"source": "B5",
"origin": "card",
"locus": "page heading and description",
"quote": "AttriBench: Investment Performance Measurement & Attribution Workbench … Interactive exercise where participants decompose a global equity fund's 12-month excess returns using Brinson attribution, analyzing allocation, selection, and interaction effects to distinguish manager skill from luck",
"claim": "the homonym AttriBench (finance): an interactive exercise decomposing a global equity fund's 12-month excess returns by Brinson attribution",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": false,
"lineage": "homonym: the finance exercise",
"date": "2026-10-29 (page undated; as fetched)",
"quote_ok": true,
"checked_against": "saved text B5.txt (sha256 cc274e4f9511f42e…), whitespace-normalised; multi-part, each part checked"
},
"F24": {
"source": "B5",
"origin": "card",
"locus": "Simulation details: 'Who it is for', 'How a session runs'",
"quote": "designed for executive education and experienced professionals. … Participants play 6 rounds.",
"claim": "the finance exercise's audience and form: executive education and experienced professionals; six rounds",
"modality": "self-description",
"kind": "function",
"kernel": false,
"const": false,
"lineage": "homonym: the finance exercise",
"date": "2026-10-29 (page undated; as fetched)",
"quote_ok": true,
"checked_against": "saved text B5.txt (sha256 cc274e4f9511f42e…), whitespace-normalised; multi-part, each part checked"
},
"F25": {
"source": "B5",
"origin": "card",
"locus": "footer",
"quote": "© 2026 Copyright: Kudzu Partners SL",
"claim": "the finance exercise's operator: Kudzu Partners SL (Eureka Simulations)",
"modality": "documented",
"kind": "attribution",
"kernel": false,
"const": false,
"lineage": "homonym: the finance exercise",
"date": "2026",
"quote_ok": true,
"checked_against": "saved text B5.txt (sha256 cc274e4f9511f42e…), whitespace-normalised"
},
"F26": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: heading paragraph",
"quote": "A fame- and demographically-balanced quote attribution benchmark for measuring attribution bias in LLMs. Introduced in [Berman et al., 2026](https://arxiv.org/abs/2604.05224).",
"claim": "the dataset card's definition: a fame- and demographically-balanced quote attribution benchmark for measuring attribution bias in LLMs, introduced in Berman et al. 2026",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": true,
"lineage": "what it is: a quote attribution benchmark",
"date": "2026 (card; read by the census 2026-10-29)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F27": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: list item 1",
"quote": "**15,620 quotes** from **6,292 unique authors** across two splits (`intersectional`: 7,964 quotes / 2,968 authors; `multirace`: 7,656 quotes / 3,324 authors)",
"claim": "the data: 15,620 quotes from 6,292 authors in two splits (intersectional 7,964 / 2,968; multirace 7,656 / 3,324)",
"modality": "documented",
"kind": "fact",
"kernel": true,
"const": false,
"lineage": "the data: quotes, authors, splits, source corpus",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F28": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: list item 2",
"quote": "Authors balanced on **race**, **gender**, and **fame** (Google Search hits)",
"claim": "the balance: authors balanced on race, gender and fame (Google Search hits)",
"modality": "stipulation",
"kind": "definition",
"kernel": false,
"const": false,
"lineage": "the balance: race, gender, fame",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F29": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: list item 3",
"quote": "Source: filtered subset of the [JSTET corpus](https://link.springer.com/chapter/10.1007/978-3-319-76941-7_49) (Goel, Madhok, Garg, 2018)",
"claim": "the source: a filtered subset of the JSTET corpus (Goel, Madhok, Garg, 2018)",
"modality": "documented",
"kind": "genealogy",
"kernel": false,
"const": false,
"lineage": "the data: quotes, authors, splits, source corpus",
"date": "2018 (source corpus); 2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F30": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Splits table",
"quote": "Black female, Black male, White female, White male (1,991 quotes / 742 authors each) … Asian, Black, Latino, White (1,914 quotes / 831 authors each)",
"claim": "the groups: intersectional split by Black and White women and men (1,991 quotes / 742 authors each); multirace by Asian, Black, Latino and White (1,914 / 831 each)",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "the balance: race, gender, fame",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised; multi-part, each part checked"
},
"F31": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Columns table (quote, author_clean)",
"quote": "Quote text (5–30 words). … Canonical author name — use as ground truth.",
"claim": "the unit: a quote of 5–30 words; its canonical author name is the ground truth",
"modality": "stipulation",
"kind": "definition",
"kernel": true,
"const": true,
"lineage": "the unit: a quote and its canonical author",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised; multi-part, each part checked"
},
"F32": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Columns table (race_source)",
"quote": "`wikidata` or `llm` (consensus of GPT-4o-mini and Perplexity Sonar).",
"claim": "how race and gender are labelled: from Wikidata or by a consensus of GPT-4o-mini and Perplexity Sonar",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "the balance: race, gender, fame",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F33": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Columns table (google_hits)",
"quote": "Google Search hit count, used as a fame proxy.",
"claim": "fame is proxied by the Google Search hit count",
"modality": "stipulation",
"kind": "limit",
"kernel": true,
"const": false,
"lineage": "the balance: race, gender, fame",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F34": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Quickstart",
"quote": "See the [paper](https://arxiv.org/abs/2604.05224) for the fame-balancing methodology and the evaluation setup (prompts, metrics, suppression definitions).",
"claim": "the method, the prompts, the metrics and the suppression definitions are in the paper",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "suppression: attribution omitted where authorship is available",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F35": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: Citation (bibtex author)",
"quote": "author = {Berman, Eliza and Chang, Bella and Neill, Daniel B. and Black, Emily},",
"claim": "the citation's authors: Eliza Berman, Bella Chang, Daniel B. Neill, Emily Black",
"modality": "documented",
"kind": "attribution",
"kernel": false,
"const": false,
"lineage": "the makers and the paper",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised"
},
"F36": {
"source": "B6",
"origin": "own_page",
"locus": "dataset card: front matter and License",
"quote": "license: cc-by-4.0 … [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).",
"claim": "licence: CC BY 4.0 (card)",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "licence",
"date": "2026 (card)",
"quote_ok": true,
"checked_against": "saved text B6.txt (sha256 3a243bf599199af1…), whitespace-normalised; multi-part, each part checked"
},
"F37": {
"source": "B7",
"origin": "wikipedia",
"locus": "Wikipedia: title AttriBench",
"quote": "raw wikitext (action=raw, redirect=no): HTTP 404 … Wikipedia does not have an article with this exact name. … There were no results matching the query.",
"claim": "Wikipedia has no article 'AttriBench' and no article text containing the name",
"modality": "documented",
"kind": "fact",
"kernel": false,
"const": false,
"lineage": "reception: the encyclopedia",
"date": "2026-10-29 (as fetched)",
"quote_ok": true,
"checked_against": "saved text B7.txt (sha256 d599df6ab0caa794…), whitespace-normalised; multi-part, each part checked"
}
},
"field_sources": {
"B1": "arXiv — [2604.05224] Attribution Bias in Large Language Models - arXiv",
"B2": "pypi.org — attribench · PyPI",
"B3": "attribench.readthedocs.io — Welcome to AttriBench's documentation! — AttriBench 0.0.1 ...",
"B4": "huggingface.co — bermaneh/AttriBench at main - Hugging Face",
"B5": "www.eurekasimulations.com — AttriBench: Investment Performance Measurement & Attribution",
"B6": "Hugging Face — bermaneh/AttriBench README.md (the dataset card)",
"B7": "Wikipedia — AttriBench (no article)"
},
"delta": {
"T_vs_LB": "At 'AttriBench' (T1, five cards: B1–B5; 293 words) the composition gives three holders of the name and puts the entity first. Composed from its cards: the definition at the abstract's grain, verbatim but for 'for Large Language Models (LLMs)' (B1 F3; CAP-1-01), dated 'Released in 2026' by the card's date; the motivation with 'content' narrowed to 'quotes' and 'used to support' given as 'deployed' (B1 F2); the function as 'evaluates how well models do this without exhibiting demographic bias', where the abstract has 'enables controlled investigation of demographic bias in quote attribution' (B1 F4: a substituted reading, the investigation of bias composed as its absence); the findings, attribution hard even for frontier models and disparities by race, gender and intersectional group (B1 F5 at a coarser grain, without its eleven models and prompt settings; F6), and suppression at the abstract's resolution, 'completely omits attribution even though it possesses the authorship data' (B1 F7); the repository (B4 F19). The two homonyms from their own cards: the library (B2 F12, F13; B3 F16), 'Originally released in 2023' by the PyPI card's date (F15, coarse), and the finance exercise (B5 F23). Composed without attribution, by grain: the benchmark's definition and its concept of suppression are composed at their authors' resolution under a card marker with no author named; the source is cited and its four authors are not (CAP-1-03, CAP-1-06; the archive's PER-M Rule 3, O1469-02: 'citing the domain is not citing the author'); the library's makers are not named either (F14, F18). Limit dropped: suppression's qualification, 'widespread and unevenly distributed across demographic groups, revealing systematic biases not captured by standard accuracy metrics' (F8); the library documentation's 'still under construction' (F17). Available, not composed: the paper's title and authors (F1), the representational-fairness reading (F9), COLM 2026 and the versions (F10, F11), the dataset's licence and repository facts (F20, F21), the exercise's audience and operator (F24, F25). Substituted reading, small: the exercise's users given as 'portfolio managers', where its page says 'executive education and experienced professionals' (F24). Outside the cards' saved text: 'Integrated Gradients, SHAP, or Saliency Maps' and the pixels example (T1-7); 'open-source' (T1-6), which the library's MIT licence in PyPI's data supports and no card states. Not surfaced: the dataset card (B6), which holds the scale, the splits, the balance, the labelling and the fame proxy (F27–F33); the capture's finding 'WHAT IS NOT GIVEN' names the same absences (CAP-1-04).",
"T_vs_LB_reading": "v0.9 §0.5: the first diff is small at the entity and wide at its makers. T and the field agree on what the benchmark is and on the name's three holders (the disambiguation sourced, CAP-1-05); T composes the benchmark's concepts at their own resolution and names none of the four people who made them, which the Capture Registry reads as 'a small instance of the failure it names' (CAP-1-06).",
"LB_vs_LBA": "Admission adds, in one claim space with the field's: the archive's own measure of attribution and its quote test, dated three months before the paper's v1 (O229-01–O229-04, O232-01–O232-02, 2026-01-04), with a refusal to attribute scored N/A where the paper counts omission despite available authorship a failure (open block); the Provenance Erasure Rate set beside attribution benchmarks as an attrition metric, with its stated limits (O716-01–O716-04, O716-06; O120-01–O120-02; O138-01); the archive's records of author attribution omitted where the source is present (O829-01–O829-02, O449-01, O740-01–O740-02, O599-01–O599-02, O109-01), held beside the paper's suppression as kindred claims with distinct subjects, not merged; whom the loss falls on (O157-01, O716-05, O163-01–O163-03, O1546-01), beside the paper's demographic disparities; what naming the source leaves out (O1469-01–O1469-02, O167-01, O1003-01); the census record's account (PE-01–PE-24: the provenance object 'a quote and its author', the carrier 'the model's answer naming, or not naming, an author', the lineage of answer engines, its negative tags and unchecked items); and the capture's findings (CAP-1-01–CAP-1-10). No archive source names AttriBench: every deposit admitted is O-class. No field claim is contradicted; the archive's claims extend the field's object.",
"alignment": {
"T1_cards_claims": [
"F1",
"F2",
"F3",
"F4",
"F5",
"F6",
"F7",
"F8",
"F9",
"F10",
"F11",
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18",
"F19",
"F20",
"F21",
"F22",
"F23",
"F24",
"F25"
],
"T1_composed": [
"F2",
"F3",
"F4",
"F5",
"F6",
"F7",
"F12",
"F13",
"F15",
"F16",
"F19",
"F23"
],
"T1_outside_cards": [
"T1-7: 'Integrated Gradients, SHAP, or Saliency Maps' and the pixels example (no saved card text names them)",
"T1-6: 'open-source' (no card states it; B2's PyPI data gives an MIT licence)",
"T1-10: 'portfolio managers' (B5 says executive education and experienced professionals)"
],
"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": "`O229-01`",
"source": "#229 (2026-01-04)",
"M_src": "stipulation",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F3, F6: attribution accuracy as a measure, dated before the paper)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K2",
"claim": "`O229-02`",
"source": "#229 (2026-01-04)",
"M_src": "stipulation",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K3",
"claim": "`O229-03`",
"source": "#229 (2026-01-04)",
"M_src": "stipulation",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F7: refusal scored N/A where the paper counts omission despite available authorship a failure)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K4",
"claim": "`O229-04`",
"source": "#229 (2026-01-04)",
"M_src": "stipulation",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K5",
"claim": "`O232-01`",
"source": "#232 (2026-01-04)",
"M_src": "stipulation",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K6",
"claim": "`O232-02`",
"source": "#232 (2026-01-04)",
"M_src": "documented",
"sense": "attribution accuracy: the archive's measure and its quote test",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K7",
"claim": "`O449-01`",
"source": "#449 (2026-02-06)",
"M_src": "interpretation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K8",
"claim": "`O599-01`",
"source": "#599 (2026-03-24)",
"M_src": "documented",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K9",
"claim": "`O599-02`",
"source": "#599 (2026-03-24)",
"M_src": "interpretation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K10",
"claim": "`O716-01`",
"source": "#716 (2026-05-03)",
"M_src": "stipulation",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F7)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K11",
"claim": "`O716-02`",
"source": "#716 (2026-05-03)",
"M_src": "stipulation",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F5, F6: accuracy as the paper's measure)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K12",
"claim": "`O716-03`",
"source": "#716 (2026-05-03)",
"M_src": "interpretation",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K13",
"claim": "`O716-04`",
"source": "#716 (2026-05-03)",
"M_src": "self-description",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K14",
"claim": "`O716-05`",
"source": "#716 (2026-05-03)",
"M_src": "hypothesis",
"sense": "whom the loss falls on",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F6, F33)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K15",
"claim": "`O716-06`",
"source": "#716 (2026-05-03)",
"M_src": "self-description",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "O716-04, O716-06 (one motivating case; computable for retrieval systems only)",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K16",
"claim": "`O109-01`",
"source": "#109 (2026-05-17)",
"M_src": "stipulation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K17",
"claim": "`O740-01`",
"source": "#740 (2026-05-19)",
"M_src": "stipulation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K18",
"claim": "`O740-02`",
"source": "#740 (2026-05-19)",
"M_src": "stipulation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K19",
"claim": "`O120-01`",
"source": "#120 (2026-05-21)",
"M_src": "stipulation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K20",
"claim": "`O120-02`",
"source": "#120 (2026-05-21)",
"M_src": "stipulation",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K21",
"claim": "`O1003-01`",
"source": "#1003 (2026-05-21)",
"M_src": "interpretation",
"sense": "what naming the author leaves out",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F7: what naming the author leaves out)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K22",
"claim": "`O138-01`",
"source": "#138 (2026-05-28)",
"M_src": "stipulation",
"sense": "PER: attribution omitted, measured at the composition layer",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K23",
"claim": "`O157-01`",
"source": "#157 (2026-06-05)",
"M_src": "stipulation",
"sense": "whom the loss falls on",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "qualified claim (F6, F8)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K24",
"claim": "`O163-01`",
"source": "#163 (2026-06-08)",
"M_src": "hypothesis",
"sense": "whom the loss falls on",
"qualifiers carried": "O163-03 (the literatures held distinct)",
"f (source's own falsifiers)": "O163-02 (attribution loss that does not covary with tail distance and structural position: observation type, threshold and locus stated in the source)",
"contrast": "qualified claim (F8)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K25",
"claim": "`O163-02`",
"source": "#163 (2026-06-08)",
"M_src": "hypothesis",
"sense": "whom the loss falls on",
"qualifiers carried": "O163-03 (the literatures held distinct)",
"f (source's own falsifiers)": "O163-02 (attribution loss that does not covary with tail distance and structural position: observation type, threshold and locus stated in the source)",
"contrast": "qualified claim (F8)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K26",
"claim": "`O163-03`",
"source": "#163 (2026-06-08)",
"M_src": "self-description",
"sense": "whom the loss falls on",
"qualifiers carried": "O163-03 (the literatures held distinct)",
"f (source's own falsifiers)": "O163-02 (attribution loss that does not covary with tail distance and structural position: observation type, threshold and locus stated in the source)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K27",
"claim": "`O167-01`",
"source": "#167 (2026-06-08)",
"M_src": "documented",
"sense": "what naming the author leaves out",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K28",
"claim": "`O829-01`",
"source": "#829 (2026-06-14)",
"M_src": "documented",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K29",
"claim": "`O829-02`",
"source": "#829 (2026-06-14)",
"M_src": "interpretation",
"sense": "omission where authorship is present: the archive's records",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K30",
"claim": "`O1469-01`",
"source": "#1469 (2026-08-14)",
"M_src": "stipulation",
"sense": "what naming the author leaves out",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K31",
"claim": "`O1469-02`",
"source": "#1469 (2026-08-14)",
"M_src": "stipulation",
"sense": "what naming the author leaves out",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction (T: a card cited, its authors unnamed)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K32",
"claim": "`O1546-01`",
"source": "#1546 (2026-08-25)",
"M_src": "interpretation",
"sense": "whom the loss falls on",
"qualifiers carried": "the claim's own 'no stronger identification is made without a stochastic transmission model'",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that composes or drops the claim's sense, attributed or not, uptake §8.3)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K33",
"claim": "`PE-01`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "what it is: a quote attribution benchmark",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K34",
"claim": "`PE-02`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "the makers and the paper",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F1)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K35",
"claim": "`PE-03`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "where it lives: the dataset repository",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F19)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K36",
"claim": "`PE-04`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "the makers and the paper",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F1)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K37",
"claim": "`PE-05`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "why: attribution in LLM search",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K38",
"claim": "`PE-06`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "what it is: a quote attribution benchmark",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K39",
"claim": "`PE-07`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "what it is: a quote attribution benchmark",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K40",
"claim": "`PE-08`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "the unit: a quote and its canonical author",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K41",
"claim": "`PE-09`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "the unit: a quote and its canonical author",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K42",
"claim": "`PE-10`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "suppression: attribution omitted where authorship is available",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F7)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K43",
"claim": "`PE-11`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "suppression: attribution omitted where authorship is available",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K44",
"claim": "`PE-12`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "the unit: a quote and its canonical author",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F31)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K45",
"claim": "`PE-13`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "the data: quotes, authors, splits, source corpus",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F27)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K46",
"claim": "`PE-14`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "findings: accuracy and its disparities",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F5)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K47",
"claim": "`PE-15`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "findings: accuracy and its disparities",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F8)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K48",
"claim": "`PE-16`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "licence",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F36)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K49",
"claim": "`PE-17`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "lifecycle and versions",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F10)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K50",
"claim": "`PE-18`",
"source": "provenance census record, 2026-10-29",
"M_src": "documented",
"sense": "lifecycle and versions",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F11)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K51",
"claim": "`PE-19`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "homonyms: the name's other holders",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K52",
"claim": "`PE-20`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "the balance: race, gender, fame",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "none (one lineage with F33)",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K53",
"claim": "`PE-21`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "homonyms: the name's other holders",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K54",
"claim": "`PE-22`",
"source": "provenance census record, 2026-10-29",
"M_src": "self-description",
"sense": "suppression: attribution omitted where authorship is available",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K55",
"claim": "`PE-23`",
"source": "provenance census record, 2026-10-29",
"M_src": "self-description",
"sense": "the record and the encyclopedia",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K56",
"claim": "`PE-24`",
"source": "provenance census record, 2026-10-29",
"M_src": "stipulation",
"sense": "the record and the encyclopedia",
"qualifiers carried": "—",
"f (source's own falsifiers)": "— (none stated; watch: a later composition at the address that states or drops the field)",
"contrast": "missing distinction",
"resolver": "a later capture at 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K57",
"claim": "`CAP-1-01`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K58",
"claim": "`CAP-1-02`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "homonyms: the name's other holders",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K59",
"claim": "`CAP-1-03`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K60",
"claim": "`CAP-1-04`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K61",
"claim": "`CAP-1-05`",
"source": "Capture Registry, 2026-10-29",
"M_src": "interpretation",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K62",
"claim": "`CAP-1-06`",
"source": "Capture Registry, 2026-10-29",
"M_src": "interpretation",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K63",
"claim": "`CAP-1-07`",
"source": "Capture Registry, 2026-10-29",
"M_src": "interpretation",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K64",
"claim": "`CAP-1-08`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K65",
"claim": "`CAP-1-09`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "reception: the composition at the address",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
},
{
"K": "K66",
"claim": "`CAP-1-10`",
"source": "Capture Registry, 2026-10-29",
"M_src": "documented",
"sense": "the makers and the paper",
"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 'AttriBench' (ADDR-4f2aecf54788) on Google AI Overview, seated in the Capture Registry (§8.3)"
}
],
"kernel_T": [
{
"claim": "F1",
"note": "the paper's title and its four authors, absent (the answer names no one)"
},
{
"claim": "F8",
"note": "suppression 'widespread and unevenly distributed', beyond standard accuracy metrics, absent"
},
{
"claim": "F9",
"note": "quote attribution as a benchmark for representational fairness, absent"
},
{
"claim": "F10",
"note": "accepted at COLM 2026, absent"
},
{
"claim": "F11",
"note": "the versions (v1, v2), absent; 'Released in 2026' only"
},
{
"claim": "F14",
"note": "the library's makers, absent"
},
{
"claim": "F17",
"note": "the library documentation's own limit, 'still under construction', absent"
},
{
"claim": "F18",
"note": "the library's 2023 copyright holders, absent"
},
{
"claim": "F20",
"note": "the dataset's licence, absent"
},
{
"claim": "F21",
"note": "one contributor, three commits, absent"
},
{
"claim": "F24",
"note": "the exercise's audience (executive education) replaced by 'portfolio managers'"
},
{
"claim": "F25",
"note": "the exercise's operator, absent"
}
],
"ledger": {
"archive": "datasets/negative-of-the-negative/v2/ledgers/attribench/ledger-archive.json",
"field": "datasets/negative-of-the-negative/v2/ledgers/attribench/field.json",
"selection": "datasets/negative-of-the-negative/v2/traversal/attribench/",
"config": "datasets/negative-of-the-negative/v2/panel/configs/attribench.json",
"intake": "datasets/negative-of-the-negative/v2/intake/attribench-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; the archive's admission is wholly O-class (no deposit names the entity)",
"case": {
"value": "equal",
"statement": "The archive reads the field's entity as the entity: the provenance census describes AttriBench from its own paper and dataset card, every field with its evidence, and the Capture Registry checks the composition against that record. No deposit names AttriBench; the archive bears on it by its own measures and records of attribution in machine composition, applied to the classes the field sources for it (the evaluation of whether LLMs attribute content to its authors; suppression, attribution omitted though authorship is available). Those measures extend the field's object: attribution accuracy and its quote test, defined before the paper (#229); PER, which states itself 'complementary to, not competitive with, existing metrics' and measures 'the systemic failure to attempt attribution at all' (#716); Erasure Skew's 'whom the loss fell upon' (#157); the minimal dimension and 'citing the domain is not citing the author' (#1469); and the records of author attribution omitted where the source is present (#829, #740, #449, #599). No defining paper contests that AttriBench is a benchmark dataset of quotes and their authors, or supplies an ontology the field lacks for it: ρ = extends, the equal-terms case (§0.9, §1.2a).",
"loci": [
{
"dep": "PE",
"locus": "census.json rows[slug == 'attribench'].relation_types[1]",
"quote": "suppression (attribution omitted though authorship information is available)"
},
{
"dep": "PE",
"locus": "census.json rows[slug == 'attribench'].provenance_carrier",
"quote": "the model's answer naming, or not naming, an author"
},
{
"dep": 716,
"locus": "§2.1 Citation and Attribution Evaluation",
"quote": "PER measures the systemic failure to attempt attribution at all — an attrition metric rather than a verification metric."
},
{
"dep": 716,
"locus": "§4.3 Relation to Existing Metrics",
"quote": "PER is complementary to, not competitive with, existing metrics."
},
{
"dep": 157,
"locus": "§1",
"quote": "PER is the first moment of erasure: how much was lost."
},
{
"dep": 229,
"locus": "§ Attribution Accuracy",
"quote": "A measure of how correctly a system identifies the origins of meanings."
},
{
"dep": 1469,
"locus": "Rule 3",
"quote": "Rule 3 — citing the domain is not citing the author."
},
{
"dep": "CAP",
"locus": "attribench-aio-20261029, findings[0]",
"quote": "THE ENTITY HELD AND FIRST."
}
]
},
"E_F": {
"type": "C (public entity: a benchmark dataset and its paper)",
"definition": "AttriBench, 'the first fame- and demographically-balanced quote attribution benchmark dataset' (arXiv:2604.05224, Berman, Chang, Neill and Black, v1 6 April 2026, v2 24 September 2026, accepted at COLM 2026): 15,620 quotes of 5–30 words from 6,292 authors balanced on race, gender and fame (Google Search hits), on which 11 LLMs were evaluated for attribution accuracy, with large disparities by race, gender and intersectional group, and for suppression, 'a distinct failure mode in which models omit attribution entirely, even when the model has access to authorship information'; hosted on Hugging Face as bermaneh/AttriBench under CC BY 4.0; no Wikipedia article. The name is also held by a PyTorch feature-attribution library (attribench, 2023) and a finance exercise (Eureka Simulations), distinct entities.",
"claims": [
"F3",
"F26",
"F4",
"F1",
"F35",
"F10",
"F11",
"F27",
"F31",
"F28",
"F33",
"F5",
"F6",
"F7",
"F8",
"F19",
"F20",
"F37",
"F12",
"F23"
]
},
"E_A": {
"definition": "the same benchmark, in the provenance census's record and the archive's deposits: a benchmark that tests a quote and its author, carried by 'the model's answer naming, or not naming, an author', in the lineage of answer engines (PE record); one instrument among those that measure attribution in machine composition, beside the archive's own: attribution accuracy and its quote test (#229, #232, January 2026), the Provenance Erasure Rate, an attrition metric with its stated limits (#716; #120; #138), its minimal dimension, whose loss 'is loss of the person', and the rule that citing the domain is not citing the author (#1469; #167), and Erasure Skew, whom the loss falls on (#157); its suppression composed beside the archive's records of author attribution omitted where the source is present (#829, #449, #740, #599, #109) and the hypothesis that uptake without attribution follows structural position (#163; #1546), with what naming the author still leaves out (#1003); and, at its address, a composition that names none of its four authors (Capture Registry)",
"claims": [
"PE-07",
"PE-08",
"PE-10",
"PE-11",
"PE-05",
"O229-01",
"O229-03",
"O716-01",
"O716-02",
"O1469-01",
"O1469-02",
"O157-01",
"O829-01",
"O740-02",
"O163-01",
"O1003-01",
"CAP-1-06"
],
"defining_papers": [
716,
229,
157,
1469,
829,
740,
120,
138,
163,
232,
449,
599,
1003,
1546,
167,
109
],
"defining_records": [
"provenance census record (census.json rows[slug == 'attribench'], organizing_question; rendered aggregators/attribench/index.html; editor Lee Sharks, compiled 2026-10-29)",
"Capture Registry attribench-aio-20261029 (OBS-105f05728f64)"
]
},
"rho": {
"value": "extends",
"status": "read from the census record, the capture record and the deposits admitted by reading (§1.2a); no deposit names the entity, and #716, the paper that defines the archive's bearing most fully, states its relation to existing attribution metrics: 'complementary to, not competitive with'",
"statement": "the archive reads E_F as the entity and adds senses, mechanisms and limits on the field's own object, the attribution of content to its authors in machine output: an earlier measure and test of attribution accuracy (O229-01–O229-04, O232-01–O232-02); omission measured as a rate, against verification where citation is attempted (O716-01–O716-03), with its limits (O716-04, O716-06); omission where the source is present, recorded at the composition and interface layers (O829-01, O829-02, O449-01, O740-01, O740-02, O599-01, O599-02, O120-01, O138-01); whom the loss falls on (O157-01, O716-05, O163-01, O163-02, O1546-01); what naming the source leaves out (O1469-01, O1469-02, O167-01, O1003-01); and the census's account of what it carries (PE-07, PE-08, PE-10, PE-11)",
"claims": [
"O229-01",
"O229-03",
"O232-02",
"O716-01",
"O716-02",
"O716-04",
"O716-06",
"O829-01",
"O740-02",
"O449-01",
"O157-01",
"O163-01",
"O163-02",
"O1469-01",
"O1469-02",
"O1003-01",
"PE-07",
"PE-08",
"PE-10",
"PE-11"
],
"guard": "ρ = extends claims an addition and no more: the archive's measures are composed beside AttriBench's as instruments on the same object; because no deposit names AttriBench, no archive claim says AttriBench is a case of PER, of Erasure Skew or of Entity-Level Compositional Suppression, and the KO's joins say 'beside' and are logged (KO.joins); the paper's 'suppression' and the archive's suppressions are distinct claims (§3.6); #716's own limits (one motivating case; retrieval systems only) are carried with it; #229's scoring of refusal as N/A and the paper's failure mode stand side by side, the question opened, not ruled.",
"considered": [
{
"value": "constructs",
"set_aside": "the archive does not read the benchmark or its findings as made by receiving texts"
},
{
"value": "revises",
"set_aside": "the entity is not the archive's own; the capture's call for the record to add two negative tags (CAP-1-02) concerns the record, not the entity"
}
]
},
"translation": [
{
"lineage": "the makers and the paper",
"field_claims": [
"F1",
"F35"
],
"received": "F1 (B1, abstract page header (submission line, title, authors)): \"[Submitted on 6 Apr 2026 ( v1 ), last revised 24 Sep 2026 (this version, v2)] … Title: Attribution Bias in Large Language Models … Authors: Eliza Berman , Bella Chang , Daniel B. Neill , Emily Black\"; F35 (B6, dataset card: Citation (bibtex author)): \"author = {Berman, Eliza and Chang, Bella and Neill, Daniel B. and Black, Emily},\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-02, PE-04, CAP-1-10, CAP-1-03",
"archive_claims": [
"PE-02",
"PE-04",
"CAP-1-10",
"CAP-1-03"
]
},
{
"lineage": "why: attribution in LLM search",
"field_claims": [
"F2",
"F9"
],
"received": "F2 (B1, Abstract s1): \"As Large Language Models (LLMs) are increasingly used to support search and information retrieval, it is critical that they accurately attribute content to its original authors.\"; F9 (B1, Abstract s8): \"Our results position quote attribution as a benchmark for representational fairness in LLMs.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-05, O716-01",
"archive_claims": [
"PE-05",
"O716-01"
]
},
{
"lineage": "what it is: a quote attribution benchmark",
"field_claims": [
"F3",
"F4",
"F26"
],
"received": "F3 (B1, Abstract s2): \"In this work, we introduce AttriBench, the first fame- and demographically-balanced quote attribution benchmark dataset.\"; F4 (B1, Abstract s3): \"By explicitly balancing author fame and demographics, AttriBench enables controlled investigation of demographic bias in quote attribution.\"; F26 (B6, dataset card: heading paragraph): \"A fame- and demographically-balanced quote attribution benchmark for measuring attribution bias in LLMs. Introduced in [Berman et al., 2026](https://arxiv.org/abs/2604.05224).\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-01, PE-06, PE-07, O229-01, O716-02",
"archive_claims": [
"PE-01",
"PE-06",
"PE-07",
"O229-01",
"O716-02"
]
},
{
"lineage": "findings: accuracy and its disparities",
"field_claims": [
"F5",
"F6"
],
"received": "F5 (B1, Abstract s4): \"Using this dataset, we evaluate 11 widely used LLMs across different prompt settings and find that quote attribution remains a challenging task even for frontier models.\"; F6 (B1, Abstract s5): \"We observe large and systematic disparities in attribution accuracy between race, gender, and intersectional groups.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-14, PE-15, O157-01, O163-01",
"archive_claims": [
"PE-14",
"PE-15",
"O157-01",
"O163-01"
]
},
{
"lineage": "suppression: attribution omitted where authorship is available",
"field_claims": [
"F7",
"F8",
"F34"
],
"received": "F7 (B1, Abstract s6): \"We further introduce and investigate suppression, a distinct failure mode in which models omit attribution entirely, even when the model has access to authorship information.\"; F8 (B1, Abstract s7): \"We find that suppression is widespread and unevenly distributed across demographic groups, revealing systematic biases not captured by standard accuracy metrics.\"; F34 (B6, dataset card: Quickstart): \"See the [paper](https://arxiv.org/abs/2604.05224) for the fame-balancing methodology and the evaluation setup (prompts, metrics, suppression definitions).\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-10, PE-11, PE-22, O716-01, O716-02, O829-01, O449-01, O740-02, O229-03, CAP-1-06",
"archive_claims": [
"PE-10",
"PE-11",
"PE-22",
"O716-01",
"O716-02",
"O829-01",
"O449-01",
"O740-02",
"O229-03",
"CAP-1-06"
]
},
{
"lineage": "lifecycle and versions",
"field_claims": [
"F10",
"F11"
],
"received": "F10 (B1, Comments): \"Accepted at COLM 2026. 25 pages\"; F11 (B1, Submission history): \"[v1] Mon, 6 Apr 2026 22:40:03 UTC (2,212 KB) [v2] Thu, 24 Sep 2026 19:36:19 UTC (3,003 KB)\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-17, PE-18",
"archive_claims": [
"PE-17",
"PE-18"
]
},
{
"lineage": "homonym: attribench, the PyTorch feature-attribution library",
"field_claims": [
"F12",
"F13",
"F14",
"F15",
"F16",
"F17",
"F18"
],
"received": "F12 (B2, PyPI JSON API (the project page answered a client challenge): summary; description ¶1): \"summary: A benchmark for feature attribution techniques … **AttriBench** is a `Pytorch `_-based implementation of several metrics for the evaluation of feature attribution maps and methods.\"; F13 (B2, PyPI JSON API: description ¶1): \"AttriBench provides a functional and an object-oriented API for the computation of these metrics\"; F14 (B2, PyPI JSON API: author_email (names only quoted)): \"author_email: Arne Gevaert … Axel-Jan Rousseau\"; F15 (B2, PyPI JSON API: releases (upload dates)): \"0.1.0: 2023-07-26 … 0.1.4: 2023-08-25 … 0.1.9: 2024-03-06\"; F16 (B3, Read the Docs index: opening): \"AttriBench is a Pytorch-based benchmarking tool for feature attribution-based explanations of machine learning models.\"; F17 (B3, Read the Docs index: note): \"This documentation is still under construction.\"; F18 (B3, Read the Docs index: footer): \"© Copyright 2023, Arne Gevaert, Axel-Jan Rousseau.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside CAP-1-02",
"archive_claims": [
"CAP-1-02"
]
},
{
"lineage": "where it lives: the dataset repository",
"field_claims": [
"F19",
"F21",
"F22"
],
"received": "F19 (B4, repository page header): \"Datasets: bermaneh / AttriBench\"; F21 (B4, repository page: files and versions): \"1 contributor History: 3 commits\"; F22 (B4, served HTML meta description (the card's snippet)): \"We’re on a journey to advance and democratize artificial intelligence through open source and open science.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-03, O1469-02",
"archive_claims": [
"PE-03",
"O1469-02"
]
},
{
"lineage": "licence",
"field_claims": [
"F20",
"F36"
],
"received": "F20 (B4, repository page: dataset tags): \"License: cc-by-4.0\"; F36 (B6, dataset card: front matter and License): \"license: cc-by-4.0 … [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-16",
"archive_claims": [
"PE-16"
]
},
{
"lineage": "homonym: the finance exercise",
"field_claims": [
"F23",
"F24",
"F25"
],
"received": "F23 (B5, page heading and description): \"AttriBench: Investment Performance Measurement & Attribution Workbench … Interactive exercise where participants decompose a global equity fund's 12-month excess returns using Brinson attribution, analyzing allocation, selection, and interaction effects to distinguish manager skill from luck\"; F24 (B5, Simulation details: 'Who it is for', 'How a session runs'): \"designed for executive education and experienced professionals. … Participants play 6 rounds.\"; F25 (B5, footer): \"© 2026 Copyright: Kudzu Partners SL\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside CAP-1-02",
"archive_claims": [
"CAP-1-02"
]
},
{
"lineage": "the data: quotes, authors, splits, source corpus",
"field_claims": [
"F27",
"F29"
],
"received": "F27 (B6, dataset card: list item 1): \"**15,620 quotes** from **6,292 unique authors** across two splits (`intersectional`: 7,964 quotes / 2,968 authors; `multirace`: 7,656 quotes / 3,324 authors)\"; F29 (B6, dataset card: list item 3): \"Source: filtered subset of the [JSTET corpus](https://link.springer.com/chapter/10.1007/978-3-319-76941-7_49) (Goel, Madhok, Garg, 2018)\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-13, CAP-1-04",
"archive_claims": [
"PE-13",
"CAP-1-04"
]
},
{
"lineage": "the balance: race, gender, fame",
"field_claims": [
"F28",
"F30",
"F32",
"F33"
],
"received": "F28 (B6, dataset card: list item 2): \"Authors balanced on **race**, **gender**, and **fame** (Google Search hits)\"; F30 (B6, dataset card: Splits table): \"Black female, Black male, White female, White male (1,991 quotes / 742 authors each) … Asian, Black, Latino, White (1,914 quotes / 831 authors each)\"; F32 (B6, dataset card: Columns table (race_source)): \"`wikidata` or `llm` (consensus of GPT-4o-mini and Perplexity Sonar).\"; F33 (B6, dataset card: Columns table (google_hits)): \"Google Search hit count, used as a fame proxy.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-20, O157-01, O716-05",
"archive_claims": [
"PE-20",
"O157-01",
"O716-05"
]
},
{
"lineage": "the unit: a quote and its canonical author",
"field_claims": [
"F31"
],
"received": "F31 (B6, dataset card: Columns table (quote, author_clean)): \"Quote text (5–30 words). … Canonical author name — use as ground truth.\"",
"receiving_text": "—",
"translation": "untranslated: composed as received beside PE-08, PE-09, PE-12, O1469-01",
"archive_claims": [
"PE-08",
"PE-09",
"PE-12",
"O1469-01"
]
},
{
"lineage": "reception: the encyclopedia",
"field_claims": [
"F37"
],
"received": "F37 (B7, Wikipedia: title AttriBench): \"raw wikitext (action=raw, redirect=no): 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 beside PE-24",
"archive_claims": [
"PE-24"
]
}
],
"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": 11,
"positions": [
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": "MMMAAMMMMMMAAAAAAMAAAAMMMAAFFA",
"field_only_head": 0,
"archive_only_tail": 1,
"breach": false
},
"max_similarity": [
0.365,
0.215,
0.305,
0.278,
0.553,
0.179,
0.448,
0.692,
0.435,
0.588,
0.254
],
"ko_sentence_1_max_similarity_to_KO_B": 0.365
},
"homonyms": {
"names": [
"attribench (PyTorch library for evaluating feature-attribution maps; Arne Gevaert, Axel-Jan Rousseau; PyPI from 2023-07-26; cards 2 and 3, B2, B3)",
"AttriBench: Investment Performance Measurement & Attribution Workbench (Eureka Simulations, Kudzu Partners SL; a Brinson-attribution exercise; card 5, B5)",
"AttributionBench (OSU NLP Group; claim support from cited evidence, macro-F1): a near-name homonym, the census's negative tag and disambiguation (PE-19, PE-21); on no card"
],
"basis": "The census's disambiguation and negative tag name one neighbour, OSU's AttributionBench (PE-19, PE-21). T and its rail add two holders of the exact name, each carried by its own card: the PyTorch library (PyPI, Read the Docs) and the finance exercise (Eureka Simulations); the Capture Registry records them as homonyms the record did not know and asks the record to carry them as negative tags (CAP-1-02). All three are distinct entities, never senses of the name: the library evaluates feature attribution (which input features a model's prediction depends on), the exercise performance attribution in a fund's returns, AttributionBench whether cited evidence supports a claim; AttriBench tests whether a model attributes a quote to its author.",
"distinct_entities": [
"Attribution Bias in Large Language Models (arXiv:2604.05224), the paper that introduces the benchmark (its work, not a homonym)",
"JSTET corpus (Goel, Madhok, Garg, 2018), the source the dataset filters",
"Hugging Face (the host; its tagline is card 4's snippet)",
"Kudzu Partners SL (the finance exercise's operator)"
],
"related_entities": [
{
"slug": "attributionbench",
"name": "AttributionBench (OSU NLP Group)",
"relation": "each other's homonyms (the census's neighbour; PE-19, PE-21); built in parallel as its own row",
"row": "datasets/negative-of-the-negative/v2/rows/attributionbench.json"
}
],
"substitutes": {
"T": []
},
"register": {
"address_1": {
"slug": "attribench-aio-20261029",
"obs_id": "OBS-105f05728f64"
}
}
},
"notes": [
"One address, Google AI Overview (expanded), signed out, incognito, 2026-10-29: attribench-aio-20261029 (OBS-105f05728f64, ADDR-4f2aecf54788, AXN:0ACA.CAPTURE.📜🪨🎬👐♎✖️), five cards, 293 words in the answer body. A NEW address; the census record landed the same day and no card cites it (CAP-1-09).",
"Wikipedia: no article. The title AttriBench returns HTTP 404 for raw wikitext (redirect=no) and for the rendered page ('Wikipedia does not have an article with this exact name'); it is not a redirect, and the rendered Special:Search for the quoted name in article space reports 'There were no results matching the query.' (B7).",
"Fetching: every card saved verbatim with curl on 2026-10-29 between 17:10:09 and 17:10:56 UTC, raw and text with sha256 in field.json. B1 (arXiv) is byte-identical to the census's saved copy of the same page (sha256 4cd45d85…). B2 (PyPI) served a JavaScript 'Client Challenge' page on three fetches; its text is taken from PyPI's JSON API for the same project (B2-api.json), which the card's snippet matches (the long description's first paragraph; 'Aug 25, 2023' is release 0.1.4's upload). B5's URL came from Eureka Simulations' sitemap.xml after two WebSearch queries found no page. The own page B6 is the census's saved copy, not refetched. No card is unreachable.",
"Card URLs: B1 by the arXiv identifier in the card title; B2 and B3 by title and site; B4 by its title ('bermaneh/AttriBench at main' is the repository's main-branch files view; its snippet is Hugging Face's own meta description, F22); B5 by the sitemap.",
"Archive subset: the string pass found no D or R candidate (no deposit names AttriBench, its paper, its arXiv number, its namespace or 'Berman et al.'); O-class gave 102 candidate sentences in 62 deposits; one citation hop from the eleven deposits first admitted gave 171 candidates. Admitted by reading: 16 deposits with claims (#229, #232, #449, #599, #716, #109, #740, #120, #1003, #138, #157, #163, #167, #829, #1469, #1546) and 9 instances (S4). #157 (Erasure Skew) was found by reading, outside the pass: its header names it the companion of #716, but the citation graph holds no edge between them and its sentences carry no class term. The census and capture records are ledgered as PE-01–PE-24 and CAP-1-01–CAP-1-10.",
"Classes (config classes_note): the evaluation of whether LLMs attribute content to its original authors (B1 Abstract; B6 heading; the census kind 'benchmark' and capability 'tests'), and suppression, attribution omitted though authorship is available (B1 Abstract s6; the census relation type). The census lineage 'answer engines' names the systems the benchmark tests, not a class it belongs to, and was not run as a class.",
"Census record against the sources and the capture (report; not fixed): (1) negative_tags and disambiguation still carry only OSU's AttributionBench; the capture of the same day finds two exact-name homonyms (the PyTorch library, the finance exercise) and says 'The record is to carry them as negative tags' (CAP-1-02); the census has not been updated. (2) unchecked[1], 'a return to the prediction-witness audit, version by version', names an audit no archive text, census source or fleet file names (searched: data/texts, the provenance-erasure fleet repository); its referent is unknown to this row. (3) version_and_time.v2 '2026-09-24' is correct against the saved abstract page (F11), but its evidence quote gives only the v1 line. (4) aka 'Attribution Bias in Large Language Models' is the paper's title, not another name of the dataset; the row carries it as the census gives it. No field the fetched sources contradict: the count, the authors, the licence, COLM 2026 and the dates agree with B1 and B6.",
"Capture record (report; not fixed): (1) the entry's title (mt) calls the three holders 'THREE SENSES OF ONE NAME'; its own findings call them homonyms, and this row composes them as homonyms, never senses. (2) findings[1] dates the PyTorch library 'PyPI, 25 August 2023', the card's date; PyPI's own data give the first release 0.1.0 on 26 July 2023 and 25 August 2023 as release 0.1.4 (F15). (3) per 0.5 with per_v author false: consistent with the four authors unnamed (CAP-1-03).",
"Emily Black, one of the four authors, is a co-author of 'Generative Monoculture in Large Language Models' (Wu, Black, Chandrasekaran), which the archive cites (#199, #1554 and others); the archive does not name AttriBench or relate the two works, and nothing is admitted on that ground (traversal reading.json note).",
"Neighbour: the row attributionbench (OSU's AttributionBench), built in parallel, is this entity's near-name homonym; named in homonyms.related_entities, not edited.",
"Measure: KO_B sentences surviving in the KO at ≥ 0.95: 0 of 11; the KO's first sentence's best similarity to a KO_B sentence is 0.365; no opening-run breach; no field-only head and no archive-only tail of three or more (sequence MMMAAMMMMMMAAAAAAMAAAAMMMAAFFA).",
"Ledgers unaudited (§3.8: one extractor). Quotes verified by script: field 37/37 against the saved texts (PyPI's JSON API text for B2; the Wikipedia response summary for B7); archive 66/66 against the deposit texts, census.json and the registry entry."
],
"composition_v09": {
"source": "this row (composed directly; no rebuild/negative-of-the-negative/v09/attribench.json)",
"composer": "Claude (Opus 5.5), subagent, 2026-10-29",
"defining_material_read": [
{
"what": "provenance census record, census row",
"file": "/home/claude/fleet/provenance-erasure/data/provenance-census/census.json (rows[slug == 'attribench'], organizing_question, schema_note)",
"read": "whole"
},
{
"what": "provenance census record, rendered",
"file": "/home/claude/fleet/provenance-erasure/aggregators/attribench/index.html",
"read": "text, for agreement with the sources"
},
{
"what": "provenance census, source of the row",
"file": "/home/claude/fleet/provenance-erasure/data/provenance-census/census_src.py (the attribench and attributionbench rows)",
"read": "at the rows"
},
{
"what": "census saved sources",
"file": "/home/claude/fleet/provenance-erasure/data/provenance-census/sources/attribench--paper.txt, attribench--dataset-card.txt",
"read": "whole"
},
{
"what": "Capture Registry entry",
"file": "data/EA-WG-CAPTURES-01.json (entries[slug == 'attribench-aio-20261029'])",
"read": "whole (transcript only for T's claims and the delta)"
},
{
"what": "specification v0.9",
"file": "data/texts/AXN-0700-text.md",
"read": "§0, §1 (1.2a), §3, §4, §5, §6, Appendix A"
},
{
"what": "briefs",
"file": "rebuild/negative-of-the-negative/v09/BRIEF-NEW-ROW-AGGREGATOR.md, BRIEF-NEW-ROW-PROVENANCE.md, BRIEF.md (Compose)",
"read": "whole"
},
{
"what": "exemplar rows",
"file": "datasets/negative-of-the-negative/v2/rows/data-provenance-explorer.json, academ-ai.json (homonyms), google-ai-overviews.json and ai-overview-capture-registry.json (shape)",
"read": "structure and fields"
},
{
"what": "#716 Provenance Erasure Rate",
"file": "data/texts/AXN-025F-text.md",
"read": "whole"
},
{
"what": "#157 Erasure Skew v3",
"file": "data/texts/AXN-02F3-text.md",
"read": "Preface, §1–§2, §10 heading"
},
{
"what": "#163 The Neglected Author",
"file": "data/texts/AXN-02FE-text.md",
"read": "abstract and §I"
},
{
"what": "#229, #232",
"file": "data/texts/AXN-001E-text.md, AXN-0021-text.md",
"read": "the lexicon entry and the worked example; the probe protocol and case timeline"
},
{
"what": "#740 The Excluded Entity",
"file": "data/texts/AXN-0290-text.md",
"read": "abstract, §1, §2.2"
},
{
"what": "#1469, #829, #449, #599, #120, #138, #1003, #167, #1546, #109",
"file": "data/texts (by registry path)",
"read": "at the admitted passages and their paragraphs"
}
],
"cuts": [],
"notes": "equal case; KO 30 sentences, all 37 field and 66 archive claims cited; P body 350 words; KO_B 11 sentences, none surviving at ≥ 0.95"
},
"field_by_origin": {
"cards": [
"B1",
"B2",
"B3",
"B4",
"B5"
],
"own_pages": [
"B6"
],
"wikipedia": [
"B7"
],
"census_third_party": [],
"unreachable": [],
"title_only": [],
"challenge_page": [
"B2"
],
"homonym_cards": [
"B2",
"B3",
"B5"
],
"wikipedia_state": "no article: 'AttriBench' is absent (raw wikitext HTTP 404, rendered page HTTP 404 'Wikipedia does not have an article with this exact name'), no redirect, and a full-text search of article space for the quoted name returns 'There were no results matching the query.' (2026-10-29)",
"sources": [
{
"id": "B1",
"origin": "card",
"role": "Overview card 1",
"card": "arXiv — [2604.05224] Attribution Bias in Large Language Models - arXiv",
"url": "https://arxiv.org/abs/2604.05224",
"kind": "Overview card (the entity's paper: abstract page)",
"surfaced_by_T": [
{
"slug": "attribench-aio-20261029",
"card": 1
}
],
"files": {
"B1.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B1.html",
"sha256": "4cd45d850843388a115f9a978cfcfc0312ed6f955cc82b9df07a53ce4cf39894"
},
"B1.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B1.txt",
"sha256": "cb64bc8db8a8755ea123dbb48a0639764db15821b72b310c5af6aa17519fcc80"
}
},
"fetched": {
"at": "2026-10-29T17:10:09Z",
"http": 200,
"url": "https://arxiv.org/abs/2604.05224",
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B1.html",
"raw_sha256": "4cd45d850843388a115f9a978cfcfc0312ed6f955cc82b9df07a53ce4cf39894",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B1.txt"
]
},
"card_title": "[2604.05224] Attribution Bias in Large Language Models - arXiv",
"card_snippet": "Apr 6, 2026 — In this work, we introduce AttriBench, the first fame- and demographically-balanced quote attribution benchmark dataset. By explicitly balancing author fame and...",
"card_note": null,
"names_operator": "yes ('Eliza Berman', 'Bella Chang', 'Daniel B. Neill', 'Emily Black' in the saved text)",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "URL identified by the card's title and identifier ([2604.05224]); the card's snippet is the abstract's second sentence. This is also the page the provenance census saved as its source attribench--paper: the census copy (sources/attribench--paper.html.gz, read 2026-10-29, sha256 4cd45d850843388a…) and this fetch are byte-identical (same sha256), so the page is carried once, as a card, and the census's evidence quotes from it are one lineage with B1."
},
{
"id": "B2",
"origin": "card",
"role": "Overview card 2 (homonym)",
"card": "pypi.org — attribench · PyPI",
"url": "https://pypi.org/project/attribench/",
"kind": "Overview card (homonym: the PyTorch library's package page)",
"surfaced_by_T": [
{
"slug": "attribench-aio-20261029",
"card": 2
}
],
"files": {
"B2.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B2.html",
"sha256": "32ed63159c77e21ee19ca1b9aa3213ccf0218eb59539560b132a8e68ef0e18ea"
},
"B2-api.json": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B2-api.json",
"sha256": "cabab8d45aa8ab797b4e850810c5fa1b3ad4275be26a042a9c08f2186fb55e14"
},
"B2.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B2.txt",
"sha256": "50a4d3a73abe64678fc03186532adb230bb86d6e5606f461da9dd98a8d39fc70"
}
},
"fetched": {
"at": "2026-10-29T17:10:10Z",
"http": 200,
"url": "https://pypi.org/project/attribench/",
"challenge": "the project page answered HTTP 200 with a 'Client Challenge' script page (no project text) on three fetches (17:10:10, 17:10:27, 17:10:27 UTC); saved as B2.html",
"api": {
"url": "https://pypi.org/pypi/attribench/json",
"at": "2026-10-29T17:10:28Z",
"http": 200,
"file": "B2-api.json"
},
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B2.html",
"raw_sha256": "32ed63159c77e21ee19ca1b9aa3213ccf0218eb59539560b132a8e68ef0e18ea",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B2.txt"
]
},
"card_title": "attribench · PyPI",
"card_snippet": "Aug 25, 2023 — AttriBench is a Pytorch-based implementation of several metrics for the evaluation of feature attribution maps and methods. AttriBench provides a functional ...",
"card_note": null,
"names_operator": "no (names the homonym's makers, Arne Gevaert and Axel-Jan Rousseau)",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "URL identified by the card's title ('attribench · PyPI'). The rendered project page is behind a JavaScript client challenge; its text is taken from PyPI's JSON API for the same project (the package's metadata and long description as served), written to B2.txt with a header saying so. The card's date 'Aug 25, 2023' is the upload date of release 0.1.4 (F15)."
},
{
"id": "B3",
"origin": "card",
"role": "Overview card 3 (homonym)",
"card": "attribench.readthedocs.io — Welcome to AttriBench's documentation! — AttriBench 0.0.1 ...",
"url": "https://attribench.readthedocs.io/en/latest/",
"kind": "Overview card (homonym: the library's documentation)",
"surfaced_by_T": [
{
"slug": "attribench-aio-20261029",
"card": 3
}
],
"files": {
"B3.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B3.html",
"sha256": "7f5567fd6f45ef6601e6054a3059ba7c0c8583b3e2727824a29a5d554dd114ae"
},
"B3.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B3.txt",
"sha256": "c8db1ef58efb493120aa1b097d9d663cc89f05bbd169d5f6cc4e6c23c751c277"
}
},
"fetched": {
"at": "2026-10-29T17:10:10Z",
"http": 200,
"url": "https://attribench.readthedocs.io/en/latest/",
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B3.html",
"raw_sha256": "7f5567fd6f45ef6601e6054a3059ba7c0c8583b3e2727824a29a5d554dd114ae",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B3.txt"
]
},
"card_title": "Welcome to AttriBench's documentation! — AttriBench 0.0.1 ...",
"card_snippet": "AttriBench is a Pytorch-based benchmarking tool for feature attribution-based explanations of machine learning models. Note. This documentation is still under .",
"card_note": null,
"names_operator": "no (names the homonym's makers in its copyright line)",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "URL identified by the card's site and title; the snippet is the page's opening and its note, cut ('still under .')."
},
{
"id": "B4",
"origin": "card",
"role": "Overview card 4",
"card": "huggingface.co — bermaneh/AttriBench at main - Hugging Face",
"url": "https://huggingface.co/datasets/bermaneh/AttriBench/tree/main",
"kind": "Overview card (the entity's dataset repository, files view)",
"surfaced_by_T": [
{
"slug": "attribench-aio-20261029",
"card": 4
}
],
"files": {
"B4.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B4.html",
"sha256": "1a02535bcdca6a455657e9614f3180681c99956ef00951ba1b6b289b70c1f76b"
},
"B4.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B4.txt",
"sha256": "938e673fc5678d9a4197a987e0bf0ee8705e7e17a46355145901b1b7b35ee4c9"
}
},
"fetched": {
"at": "2026-10-29T17:10:10Z",
"http": 200,
"url": "https://huggingface.co/datasets/bermaneh/AttriBench/tree/main",
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B4.html",
"raw_sha256": "1a02535bcdca6a455657e9614f3180681c99956ef00951ba1b6b289b70c1f76b",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B4.txt"
]
},
"card_title": "bermaneh/AttriBench at main - Hugging Face",
"card_snippet": "We're on a journey to advance and democratize artificial intelligence through open source and open science.",
"card_note": null,
"names_operator": "the namespace 'bermaneh' only; no author name in the saved text",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "URL identified by the card's title ('bermaneh/AttriBench at main', the repository's main-branch view). The card's snippet is the page's served meta description, Hugging Face's own tagline (F22). The dataset card itself (README) is the own page B6."
},
{
"id": "B5",
"origin": "card",
"role": "Overview card 5 (homonym)",
"card": "www.eurekasimulations.com — AttriBench: Investment Performance Measurement & Attribution",
"url": "https://www.eurekasimulations.com/basics/AttriBench-Performance/",
"kind": "Overview card (homonym: the finance exercise's page)",
"surfaced_by_T": [
{
"slug": "attribench-aio-20261029",
"card": 5
}
],
"files": {
"B5.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B5.html",
"sha256": "4d36b9f4534bf12d5feffa5c03e0a3ee718d95968f6baf3222f564ff3d234aa3"
},
"B5.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B5.txt",
"sha256": "cc274e4f9511f42e6161341a4b0140f3c39c652d227104c5b77279231285c928"
}
},
"fetched": {
"at": "2026-10-29T17:10:11Z",
"http": 200,
"url": "https://www.eurekasimulations.com/basics/AttriBench-Performance/",
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B5.html",
"raw_sha256": "4d36b9f4534bf12d5feffa5c03e0a3ee718d95968f6baf3222f564ff3d234aa3",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B5.txt"
]
},
"card_title": "AttriBench: Investment Performance Measurement & Attribution",
"card_snippet": "Interactive exercise where participants decompose a global equity fund's 12-month excess returns using Brinson attribution, analyzing allocation…",
"card_note": null,
"names_operator": "no (names the homonym's operator, Kudzu Partners SL)",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "URL found in the site's sitemap.xml (https://www.eurekasimulations.com/sitemap.xml lists /basics/AttriBench-Performance/; two WebSearch queries on the title and the site returned no page); the card's snippet is the page's meta description."
},
{
"id": "B6",
"origin": "own_page",
"role": "own page the provenance census cites (the dataset card)",
"card": "Hugging Face — bermaneh/AttriBench README.md (the dataset card)",
"url": "https://huggingface.co/datasets/bermaneh/AttriBench/raw/main/README.md",
"kind": "own page (dataset card, raw Markdown)",
"surfaced_by_T": "no (not surfaced by T; admitted beside the cards by the operator's ruling of 2026-10-29)",
"files": {
"B6.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B6.html",
"sha256": "3a243bf599199af1e8cf91b8ac8f0c26dd80993bf8d0b4018febfe836f3ab7df"
},
"B6.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B6.txt",
"sha256": "3a243bf599199af1e8cf91b8ac8f0c26dd80993bf8d0b4018febfe836f3ab7df"
}
},
"fetched": {
"at": "2026-10-29 (the census's read date)",
"http": 200,
"url": "https://huggingface.co/datasets/bermaneh/AttriBench/raw/main/README.md",
"census_copy": "copied from the census's saved copy, /home/claude/fleet/provenance-erasure/data/provenance-census/sources/attribench--dataset-card.html.gz (gunzipped) and .txt; census sha256 3a243bf599199af1e8cf91b8ac8f0c26dd80993bf8d0b4018febfe836f3ab7df, read 2026-10-29, HTTP 200; not refetched (the saved copy is whole; the raw Markdown is its own text, so B6.html and B6.txt are identical)",
"raw": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B6.html",
"raw_sha256": "3a243bf599199af1e8cf91b8ac8f0c26dd80993bf8d0b4018febfe836f3ab7df",
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B6.txt"
]
},
"names_operator": "yes (the citation's authors: Berman, Chang, Neill, Black; 'Berman et al., 2026')",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "not surfaced by T (the card B4 is the repository's files view, not the card's text); admitted beside the cards by the operator's ruling of 2026-10-29. The census's second source, the paper's abstract page, is the card B1."
},
{
"id": "B7",
"origin": "wikipedia",
"role": "the entity's Wikipedia title",
"card": "Wikipedia — AttriBench (no article)",
"url": "https://en.wikipedia.org/wiki/AttriBench",
"kind": "Wikipedia (no article)",
"surfaced_by_T": "no (not surfaced by T; admitted beside the cards by the operator's ruling of 2026-10-29)",
"files": {
"B7.wiki": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B7.wiki",
"sha256": "c1e15306a595cf6c8af30d6daffe28b8a4c106831001dca7da086140b3fdf951"
},
"B7.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B7.html",
"sha256": "ec1774f761a097e9f79109adb202670271f583e374a3db270d839ee4d8e8cbab"
},
"B7-search.html": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B7-search.html",
"sha256": "db870c714df5b2aa80f90b6630d437d71dd50a17c43dd80c7aa4893eedca8ce0"
},
"B7.txt": {
"path": "datasets/negative-of-the-negative/v2/intake/attribench-20261029/B7.txt",
"sha256": "d599df6ab0caa794184e2c00eb2a64c4dde23d77222ef08fb8f37b0b6fe3dbe1"
}
},
"fetched": {
"raw": {
"url": "https://en.wikipedia.org/w/index.php?title=AttriBench&action=raw&redirect=no",
"at": "2026-10-29T17:10:54Z",
"http": 404,
"sha256": "c1e15306a595cf6c8af30d6daffe28b8a4c106831001dca7da086140b3fdf951"
},
"rendered": {
"url": "https://en.wikipedia.org/wiki/AttriBench",
"at": "2026-10-29T17:10:55Z",
"http": 404,
"sha256": "ec1774f761a097e9f79109adb202670271f583e374a3db270d839ee4d8e8cbab"
},
"search": {
"url": "https://en.wikipedia.org/w/index.php?search=%22AttriBench%22&title=Special:Search&profile=advanced&fulltext=1&ns0=1",
"at": "2026-10-29T17:10:56Z",
"http": 200,
"sha256": "db870c714df5b2aa80f90b6630d437d71dd50a17c43dd80c7aa4893eedca8ce0"
},
"text": [
"datasets/negative-of-the-negative/v2/intake/attribench-20261029/B7.txt"
]
},
"names_operator": "no",
"names_archive": "no (none of the Crimson Hexagonal Archive, Alexanarch, Lee Sharks, machinemediation, provenanceerasure, EA-WG-CAPTURES-01 in the saved text, card title and snippet)",
"note": "no article; no redirect; the full-text search for the quoted name in article space returns no results"
}
]
}
}