{
 "spxi_instrument": {
  "method": "0.9",
  "name": "Capture Registry (longitudinal observation)",
  "level": 0,
  "level_name": "Measurement",
  "url": "https://www.spxi.dev/methods/#m-0-9",
  "inventory": "https://www.spxi.dev/methods/",
  "statement": "SPXI instrument: method 0.9 of 87, level 0 (Measurement), Capture Registry (longitudinal observation).",
  "also": []
 },
 "slug": "attributionbench-aio-20261029",
 "date": "2026-10-29",
 "surface": "Google AI Overview",
 "surface_basis": "Operator default of 2026-10-01 (runs begin in the Overview unless he says AI Mode); the 'AI Mode Conversation' paste is the Overview expanded (operator, 2026-10-18).",
 "surfaces": [
  "Google AI Overview"
 ],
 "auth": "signed out, incognito",
 "auth_basis": "Operator default of 2026-10-14.",
 "ev": "paste",
 "mt": "ATTRIBUTIONBENCH ON THE OVERVIEW — OSU'S BENCHMARK EXACT: SEVEN SOURCE DATASETS, THE BINARY TASK, ABOUT 80% MACRO-F1; KEPT APART FROM ATTRIBENCH",
 "match": "BROAD MATCH",
 "match_note": null,
 "s": "Machine Reception",
 "q": "AttributionBench",
 "q_kind": "the name of an entity of the provenance meta-aggregator (AttributionBench; record https://www.provenanceerasure.org/aggregators/attributionbench/, compiled 2026-10-29), the bare entity string supplied at the operator's request ('strings on the bare entity, no follow questions') and run by him the same day, to supply the composition layer's account of the entity for its record and its /non row. A baseline: the record landed on provenanceerasure.org the same day (commit 8a9f153) and no card cites it. NEW address.",
 "cites": 5,
 "cite_list": [
  {
   "n": 1,
   "site": "arXiv",
   "rel": "third_party",
   "url": null,
   "title": "[2402.15089] AttributionBench: How Hard is Automatic Attribution Evaluation?",
   "snip": "AttributionBench: How Hard is Automatic Attribution Evaluation? Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun View a PDF of the paper titled AttributionBench: How Ha...",
   "note": null
  },
  {
   "n": 2,
   "site": "GitHub",
   "rel": "third_party",
   "url": null,
   "title": "GitHub - OSU-NLP-Group/AttributionBench: [ACL'24 Findings] AttributionBench: How Hard is Automatic Attribution Evaluation? · GitHub",
   "snip": "Paper: AttributionBench: How Hard is Automatic Attribution Evaluation? (Accepted to ACL 2024 Findings). Dataset: Unified testbed for evaluating attribution eval...",
   "note": null
  },
  {
   "n": 3,
   "site": "GitHub Pages documentation",
   "rel": "third_party",
   "url": null,
   "title": "AttributionBench: How Hard is Automatic Attribution Evaluation?",
   "snip": "Overview: AttributionBench is a comprehensive benchmark compiled from existing attribution datasets (ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HA...",
   "note": null
  },
  {
   "n": 4,
   "site": "Hugging Face",
   "rel": "third_party",
   "url": null,
   "title": "osunlp/AttributionBench · Datasets at Hugging Face",
   "snip": "Dataset: osunlp/AttributionBench (arXiv: 2402.15089, created by Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun). Task/Modality: Text Classification, Text modality, En...",
   "note": null
  },
  {
   "n": 5,
   "site": "ACL Anthology",
   "rel": "third_party",
   "url": null,
   "title": "AttributionBench: How Hard is Automatic Attribution Evaluation?",
   "snip": "Abstract Modern generative search engines enhance the reliability of large language model (LLM) responses by providing cited evidence. However, evaluating the a...",
   "note": null
  }
 ],
 "archive_controlled_cites": 0,
 "transcript": "[Google AI Overview at the query, signed out and incognito (operator defaults of 2026-10-01 and 2026-10-14), 2026-10-29. The paste is the Overview expanded (the 'AI Mode Conversation' view, with the source cards that follow the answer); the 'You said' line carries the query twice, run together, and is given once here. Inline links are Google redirect URLs, kept as pasted.]\n\n[QUERENT] AttributionBench\n\n[SYSTEM] [AttributionBench](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg) is a comprehensive research benchmark designed to evaluate automatic attribution and citation faithfulness in Large Language Model (LLM) and generative search engine responses. [[1](https://www.google.com/goto?url=CAESWAHrOzAVmqv2rMzE2fwWGexH4ZP2TmOPFx1x4yOBvoEDc5-2uk21FgRGxqSTveJLXNIK9i_jG2EfxOjIDd-0iwHArcW44frlGDmFVP3i31HMP1TN30TmK00), [2](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\nOverview & Purpose\n\n* Unified Testbed: It compiles multiple existing attribution datasets—including ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HAGRID, and AttrEval-GenSearch—into a single standardized evaluation format. [[1](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ), [2](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\n* Core Task: It measures whether every claim generated by an LLM is fully and accurately supported by its cited reference evidence (classified as `attributable` or `not attributable`). [[1](https://www.google.com/goto?url=CAESZwHrOzAVpZlERTl75IWAAC9-thaAFB3DX_b_T81TUeWKlJZW4_YdoOcnA7wZoHkgXY3Zhbc2_YPw_zVIs-_knsuf7b3o6gPpZYjI3KpLPpPMDvadBfybX5T3dwU_ZEBc_-QNEra_4O0), [2](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ)]\n* Dataset Availability: The dataset and code are open-sourced via the [Hugging Face AttributionBench Dataset](https://www.google.com/goto?url=CAESbwHrOzAVNHLVxrI2cQf_XAnHqx_nRxad9fjvo47I43P_6udYsZree37Tf4h0ThugpBuyHFG3GeSlRkiVwIcWexDwXVFexctl11F4y0phY62RZFAdTEAa6oiODlBMLLZghaBGvVilCCYxJBbUecAC_Q) and the [GitHub Repository](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ). [[1](https://www.google.com/goto?url=CAESbwHrOzAVNHLVxrI2cQf_XAnHqx_nRxad9fjvo47I43P_6udYsZree37Tf4h0ThugpBuyHFG3GeSlRkiVwIcWexDwXVFexctl11F4y0phY62RZFAdTEAa6oiODlBMLLZghaBGvVilCCYxJBbUecAC_Q), [2](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ)]\n\nKey Findings\n\n* High Difficulty for LLMs: Even cutting-edge models like GPT-4 and fine-tuned GPT-3.5 face significant limitations, reaching only around 80% macro-F1 score on binary attribution classification tasks (and dropping to 60–70% or lower on complex, long-evidence, or out-of-domain datasets). [[1](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg), [2](https://www.google.com/goto?url=CAESWAHrOzAVmqv2rMzE2fwWGexH4ZP2TmOPFx1x4yOBvoEDc5-2uk21FgRGxqSTveJLXNIK9i_jG2EfxOjIDd-0iwHArcW44frlGDmFVP3i31HMP1TN30TmK00)]\n* Error Breakdown: Analysis of model failures shows that over 66% of errors stem from fine-grained information insensitivity (missing subtle nuances in the text), while nearly 27% derive from mismatches between what the model has access to and human-level context understanding. [[1](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\n\nWould you like to explore how to load AttributionBench via Hugging Face or learn more about the specific models and fine-tuning scripts used in the paper?\narXiv\n[2402.15089] AttributionBench: How Hard is Automatic Attribution Evaluation?\nAttributionBench: How Hard is Automatic Attribution Evaluation? Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun View a PDF of the paper titled AttributionBench: How Ha...\nGitHub\nGitHub - OSU-NLP-Group/AttributionBench: [ACL'24 Findings] AttributionBench: How Hard is Automatic Attribution Evaluation? · GitHub\nPaper: AttributionBench: How Hard is Automatic Attribution Evaluation? (Accepted to ACL 2024 Findings). Dataset: Unified testbed for evaluating attribution eval...\nGitHub Pages documentation\nAttributionBench: How Hard is Automatic Attribution Evaluation?\nOverview: AttributionBench is a comprehensive benchmark compiled from existing attribution datasets (ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HA...\nHugging Face\nosunlp/AttributionBench · Datasets at Hugging Face\nDataset: osunlp/AttributionBench (arXiv: 2402.15089, created by Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun). Task/Modality: Text Classification, Text modality, En...\nACL Anthology\nAttributionBench: How Hard is Automatic Attribution Evaluation?\nAbstract Modern generative search engines enhance the reliability of large language model (LLM) responses by providing cited evidence. However, evaluating the a...",
 "transcript_raw": "[AttributionBench](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg) is a comprehensive research benchmark designed to evaluate automatic attribution and citation faithfulness in Large Language Model (LLM) and generative search engine responses. [[1](https://www.google.com/goto?url=CAESWAHrOzAVmqv2rMzE2fwWGexH4ZP2TmOPFx1x4yOBvoEDc5-2uk21FgRGxqSTveJLXNIK9i_jG2EfxOjIDd-0iwHArcW44frlGDmFVP3i31HMP1TN30TmK00), [2](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\nOverview & Purpose\n\n* Unified Testbed: It compiles multiple existing attribution datasets—including ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HAGRID, and AttrEval-GenSearch—into a single standardized evaluation format. [[1](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ), [2](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\n* Core Task: It measures whether every claim generated by an LLM is fully and accurately supported by its cited reference evidence (classified as `attributable` or `not attributable`). [[1](https://www.google.com/goto?url=CAESZwHrOzAVpZlERTl75IWAAC9-thaAFB3DX_b_T81TUeWKlJZW4_YdoOcnA7wZoHkgXY3Zhbc2_YPw_zVIs-_knsuf7b3o6gPpZYjI3KpLPpPMDvadBfybX5T3dwU_ZEBc_-QNEra_4O0), [2](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ)]\n* Dataset Availability: The dataset and code are open-sourced via the [Hugging Face AttributionBench Dataset](https://www.google.com/goto?url=CAESbwHrOzAVNHLVxrI2cQf_XAnHqx_nRxad9fjvo47I43P_6udYsZree37Tf4h0ThugpBuyHFG3GeSlRkiVwIcWexDwXVFexctl11F4y0phY62RZFAdTEAa6oiODlBMLLZghaBGvVilCCYxJBbUecAC_Q) and the [GitHub Repository](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ). [[1](https://www.google.com/goto?url=CAESbwHrOzAVNHLVxrI2cQf_XAnHqx_nRxad9fjvo47I43P_6udYsZree37Tf4h0ThugpBuyHFG3GeSlRkiVwIcWexDwXVFexctl11F4y0phY62RZFAdTEAa6oiODlBMLLZghaBGvVilCCYxJBbUecAC_Q), [2](https://www.google.com/goto?url=CAESaQHrOzAVgms_ESOyHhe1D0kETAdvLhlfTSZSqGbiiS0TIUDdgAqEf3qpe-ByJYOTGx1ji1SsDDlVDzXKZzBQYdVtbZZ0FbJ04coryqRDcGSfdgwimQIVGpk49t8HHDc6nI43aLp_ka5ejQ)]\n\nKey Findings\n\n* High Difficulty for LLMs: Even cutting-edge models like GPT-4 and fine-tuned GPT-3.5 face significant limitations, reaching only around 80% macro-F1 score on binary attribution classification tasks (and dropping to 60–70% or lower on complex, long-evidence, or out-of-domain datasets). [[1](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg), [2](https://www.google.com/goto?url=CAESWAHrOzAVmqv2rMzE2fwWGexH4ZP2TmOPFx1x4yOBvoEDc5-2uk21FgRGxqSTveJLXNIK9i_jG2EfxOjIDd-0iwHArcW44frlGDmFVP3i31HMP1TN30TmK00)]\n* Error Breakdown: Analysis of model failures shows that over 66% of errors stem from fine-grained information insensitivity (missing subtle nuances in the text), while nearly 27% derive from mismatches between what the model has access to and human-level context understanding. [[1](https://www.google.com/goto?url=CAESaQHrOzAVU01FbKjqKLFYWTaNiZYplQ8UQ-cTMSNBjPF8nqEl2y9--9eB-WyrqaP046sOGs1WcXCz5z-vjxbFnIGx2wjxuo7Qo5Q8WmHtkKQlAa5EjDFmUSFi31CF9mMLgtKU2HaYdleBfg)]\n\nWould you like to explore how to load AttributionBench via Hugging Face or learn more about the specific models and fine-tuning scripts used in the paper?\narXiv\n[2402.15089] AttributionBench: How Hard is Automatic Attribution Evaluation?\nAttributionBench: How Hard is Automatic Attribution Evaluation? Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun View a PDF of the paper titled AttributionBench: How Ha...\nGitHub\nGitHub - OSU-NLP-Group/AttributionBench: [ACL'24 Findings] AttributionBench: How Hard is Automatic Attribution Evaluation? · GitHub\nPaper: AttributionBench: How Hard is Automatic Attribution Evaluation? (Accepted to ACL 2024 Findings). Dataset: Unified testbed for evaluating attribution eval...\nGitHub Pages documentation\nAttributionBench: How Hard is Automatic Attribution Evaluation?\nOverview: AttributionBench is a comprehensive benchmark compiled from existing attribution datasets (ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HA...\nHugging Face\nosunlp/AttributionBench · Datasets at Hugging Face\nDataset: osunlp/AttributionBench (arXiv: 2402.15089, created by Yifei Li, Xiang Yue, Zeyi Liao, Huan Sun). Task/Modality: Text Classification, Text modality, En...\nACL Anthology\nAttributionBench: How Hard is Automatic Attribution Evaluation?\nAbstract Modern generative search engines enhance the reliability of large language model (LLM) responses by providing cited evidence. However, evaluating the a...",
 "transcript_class": "CAPTURE-TIME VERBATIM RECORD — paste of the Overview expanded, with its source cards",
 "transcript_complete": "complete as supplied",
 "transcript_read": "READ IN FULL 2026-10-29",
 "per": 0.5,
 "per_corrected": null,
 "per_v": {
  "author": false,
  "inst": false,
  "id": true,
  "src": true
 },
 "per_note": "No person or institution in the answer; the arXiv identifier is in a card; the paper, repository and dataset cited.",
 "availability": null,
 "per_dim": null,
 "per_dim_v1_0": null,
 "sf": null,
 "sf_derived": null,
 "reading": "The innocent reading first. The Overview identifies the benchmark, what it compiles, what it asks and how hard models find it, all from the paper and its repositories.",
 "analysis": "An attribution benchmark is composed at full grain from its own paper and kept distinct from its homonym, and its makers are left in the cards. The record's negative tag, that support by evidence is not credit to the original author, is the line between this entity and the next.",
 "d": "ATTRIBUTIONBENCH, BY NAME, ON GOOGLE'S AI OVERVIEW, SIGNED OUT, 2026-10-29: 'a comprehensive research benchmark designed to evaluate automatic attribution and citation faithfulness in Large Language Model (LLM) and generative search engine responses.' A unified testbed of seven datasets (ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HAGRID, AttrEval-GenSearch), the task (is every claim fully supported by its cited evidence: attributable or not), the data on Hugging Face and GitHub; findings (about 80% macro-F1 for GPT-4 and fine-tuned GPT-3.5, 60–70% on harder sets; over 66% of errors from fine-grained information insensitivity, nearly 27% from a mismatch of access and context). Five cards: arXiv 2402.15089, the OSU-NLP-Group repository, the project page, the Hugging Face dataset, the ACL Anthology.",
 "d_full": "ATTRIBUTIONBENCH, BY NAME, ON GOOGLE'S AI OVERVIEW, SIGNED OUT, 2026-10-29: 'a comprehensive research benchmark designed to evaluate automatic attribution and citation faithfulness in Large Language Model (LLM) and generative search engine responses.' A unified testbed of seven datasets (ExpertQA, Stanford-GenSearch, AttributedQA, LFQA, BEGIN, HAGRID, AttrEval-GenSearch), the task (is every claim fully supported by its cited evidence: attributable or not), the data on Hugging Face and GitHub; findings (about 80% macro-F1 for GPT-4 and fine-tuned GPT-3.5, 60–70% on harder sets; over 66% of errors from fine-grained information insensitivity, nearly 27% from a mismatch of access and context). Five cards: arXiv 2402.15089, the OSU-NLP-Group repository, the project page, the Hugging Face dataset, the ACL Anthology.",
 "d_truncated": false,
 "links": [
  {
   "url": "https://www.alexanarch.org/captures/attributionbench-aio-20261029/",
   "authority": "canonical",
   "note": "the capture's own record page; cite this form"
  },
  {
   "url": "https://www.alexanarch.org/captures/#attributionbench-aio-20261029",
   "authority": "gallery",
   "note": "the canonical gallery, anchored by slug"
  },
  {
   "url": "https://www.godkinggoogle.com/captures/#attributionbench-aio-20261029",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  },
  {
   "url": "https://www.leesharks.com/captures/#attributionbench-aio-20261029",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  },
  {
   "url": "https://www.machinemediation.org/captures/#attributionbench-aio-20261029",
   "authority": "mirror",
   "note": "a window that renders from the archive's registry; may lag a deploy"
  }
 ],
 "imgs": [],
 "img_urls": [],
 "cite": "https://www.alexanarch.org/captures/attributionbench-aio-20261029/",
 "citable_unit": "address — the exact issued string on one surface, per the Surface Rule (MANUS, 2026-08-15)",
 "addr_id": "ADDR-46cc2d945075",
 "obs_id": "OBS-90d2b25bd00a",
 "n_observations": 1,
 "observations": [],
 "dates": [
  "2026-10-29"
 ],
 "defects": [],
 "findings": [
  "THE ENTITY EXACT. The task ('whether every claim within the generated responses is fully supported by its cited evidence') and the 80% macro-F1 are the record's quotes from the project page. Checked 2026-10-29 against the provenance census record of the same day and the sources it saved.",
  "KEPT APART. The composition never mentions AttriBench, the quote-attribution benchmark with a near name; the record's negative tag ('NOT AttriBench') is respected by silence.",
  "THE AUTHORS IN THE CARDS ONLY. Yifei Li, Xiang Yue, Zeyi Liao and Huan Sun are named in the arXiv and Hugging Face cards; the answer names neither them nor the OSU NLP Group.",
  "THE ERROR SHARES. The 66% and 27% figures are the paper's error analysis; on the project page the record saved ('Over 66%', 'About 26.8%'); the record does not carry them. (Repaired 2026-10-29: the finding had called them unchecked.)"
 ],
 "series": null,
 "other_slugs": null,
 "collisions": null,
 "oq": null,
 "rounds": null,
 "turns": 1,
 "rerun": null,
 "rerun_alt": null,
 "heteronym": null,
 "model_attribution": null,
 "operator_disclosure": null,
 "longitudinal_priors": null,
 "longitudinal_successors": null,
 "related_deposits": [],
 "intended": null,
 "originator": {
  "name": "OSU NLP Group (The Ohio State University)",
  "relation": "external",
  "entity_type": "dataset",
  "spxi_treatment": "partial",
  "basis": "Not the archive's: AttributionBench; its operator, as the record gives it: OSU NLP Group (The Ohio State University). Formalized by the provenance meta-aggregator record of 2026-10-29 (https://www.provenanceerasure.org/aggregators/attributionbench/), editor Lee Sharks; the record's fields carry verbatim quotes from the sources it saved with sha256, each with its basis (the operator's own page or a third-party account). Recorded 2026-10-29; repaired the same day."
 },
 "notes": {
  "prompts": "One query. Sent by the operator on 2026-10-29 with thirteen others, one per entity of the provenance meta-aggregator, from the strings supplied at his request ('shoot me those search strings. strings on the bare entity, no follow questions'); pasted in one message separated by '$'. The paste was lost to a context compaction of this session and re-pasted by the operator the same day; the second paste is the one seated."
 },
 "notice": null,
 "axn_root": "AXN:0ACB.CAPTURE",
 "axn": "AXN:0ACB.CAPTURE.🎇🐝🧡🌓🚀🔐",
 "record_url": "https://www.alexanarch.org/captures/attributionbench-aio-20261029/"
}