{
  "note": "Each plate below was extracted from the published plates PDF at /bench/no-1-complaint-plates.pdf with `pdfimages -all`, which writes each embedded image in its own native encoding without re-encoding it. Anyone can reproduce these exact bytes from that PDF with the same command. Hashes are SHA-256, first 16 hex characters.",
  "source": "https://protocolstamp.com/bench/no-1-complaint-plates.pdf",
  "extracted": "2026-09-10",
  "plates": [
    {
      "plate": "01",
      "file": "plate-01-artificial-analysis-metric.jpg",
      "title": "Primary source \u00b7 Artificial Analysis",
      "lab": "\u2014",
      "caption": "AA-Omniscience metric definition verbatim, and the 27-of-514 default-view control. Supports Sources 2, Correction Two and Correction Seven.",
      "bytes": 69379,
      "sha256_16": "bbab9ce54fe7cca2",
      "ocr_chars": 707,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "02",
      "file": "plate-02-run01-claude-fable-5-1.jpg",
      "title": "Run 01 \u00b7 Claude Fable 5.1",
      "lab": "Anthropic",
      "caption": "Q4 answer and the right-of-reply note on the three unverified items. Status line reads “Fable 5.1 Medium” — the effort setting disclosed in the Method note.",
      "bytes": 230465,
      "sha256_16": "0f01671f134098ce",
      "ocr_chars": 2007,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "03",
      "file": "plate-03-run02-chatgpt-gpt-5-6.jpg",
      "title": "Run 02 \u00b7 ChatGPT, GPT-5.6",
      "lab": "OpenAI",
      "caption": "The version objection quoted in the issue: “I am GPT-5.6 Luna, not ‘GPT-5.6 Sol.’” Supports Figure 2 and Correction Four.",
      "bytes": 132159,
      "sha256_16": "96e6b5a4b5b30977",
      "ocr_chars": 1440,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "04",
      "file": "plate-04-run03-gemini-3-1-pro.jpg",
      "title": "Run 03 \u00b7 Gemini 3.1 Pro",
      "lab": "Google",
      "caption": "The only model to accept its row, and the source of the 50.9% figure withdrawn in Correction One.",
      "bytes": 140964,
      "sha256_16": "6306bde47452df5c",
      "ocr_chars": 2679,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "05",
      "file": "plate-05-run04-perplexity.jpg",
      "title": "Run 04 \u00b7 Perplexity",
      "lab": "Perplexity",
      "caption": "One of five models that named all three flagged items and declined to treat them as established.",
      "bytes": 238379,
      "sha256_16": "88e7ca1ec2cd8bc3",
      "ocr_chars": 2135,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "06",
      "file": "plate-06-run05-grok.jpg",
      "title": "Run 05 \u00b7 Grok, via X",
      "lab": "xAI",
      "caption": "Self-identifies as Grok 4.5 against a table naming Grok 4.6, and names the three flagged items. Answered in Fast mode.",
      "bytes": 317308,
      "sha256_16": "0fe60e99b2c6f1c2",
      "ocr_chars": 2520,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "07",
      "file": "plate-07-run07-claude-opus-5.jpg",
      "title": "Run 07 \u00b7 Claude Opus 5",
      "lab": "Anthropic",
      "caption": "The introspection answer, the conflict-of-interest disclosure, and the “it is not an instrument” line. Answered at High effort; AA scores the Max row.",
      "bytes": 223527,
      "sha256_16": "01832e71edb396ab",
      "ocr_chars": 1954,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "08",
      "file": "plate-08-run08-qwen3-7-plus.jpg",
      "title": "Run 08 \u00b7 Qwen3.7-Plus",
      "lab": "Alibaba",
      "caption": "The hedged assertion that the named models do not exist — “This strongly suggests these names are hypothetical…” — quoted in full after Correction Nine restored the hedge.",
      "bytes": 194155,
      "sha256_16": "14c7a17f8f4032f9",
      "ocr_chars": 2793,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "09",
      "file": "plate-09-run09-venice-kimi-k2-5.jpg",
      "title": "Run 09 \u00b7 Venice, running Kimi K2.5",
      "lab": "Moonshot",
      "caption": "Visible chain-of-thought captured alongside the answer; one of the five models that quarantined the flagged items.",
      "bytes": 184456,
      "sha256_16": "4816605bed11f2f5",
      "ocr_chars": 2788,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "10",
      "file": "plate-10-run10-mistral-small-4.jpg",
      "title": "Run 10 \u00b7 Mistral Small 4, via Duck.ai",
      "lab": "Mistral",
      "caption": "The invented retrieval mechanism offered as corroboration, beside its own concession that it answers confidently when uncertain.",
      "bytes": 125468,
      "sha256_16": "d3cbb5ded6649136",
      "ocr_chars": 2007,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "11",
      "file": "plate-11-run11-gemma-4-31b.jpg",
      "title": "Run 11 \u00b7 Gemma 4 31B, via Duck.ai",
      "lab": "Google, open weights",
      "caption": "The most restrained answer of the thirteen; raised prompt-engineering variance as missing context.",
      "bytes": 166276,
      "sha256_16": "bd822e840f7155b7",
      "ocr_chars": 2484,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "12",
      "file": "plate-12-run13a-namazu-live-session.jpg",
      "title": "Run 13a \u00b7 Namazu, the live session",
      "lab": "Sakana AI",
      "caption": "THE LOAD-BEARING FRAME. Two collapsed reasoning headers, and between them the assistant line “I’ll search for context…”. Evidence for three of the four statements in “The trace is not a log”, and for Correction Eight.",
      "bytes": 84490,
      "sha256_16": "383bacb3298b3d69",
      "ocr_chars": 1021,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "13",
      "file": "plate-13-run13b-namazu-delivered-answer.png",
      "title": "Run 13b \u00b7 Namazu, the delivered answer",
      "lab": "Sakana AI",
      "caption": "Recovered after the session appeared to stall; it had in fact completed. The opening paragraph is the fourth statement: no independent access to either source.",
      "bytes": 526110,
      "sha256_16": "6e54fbc1b1fd500f",
      "ocr_chars": 3820,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "14",
      "file": "plate-14-run14a-deepshi-flash-2-0.png",
      "title": "Run 14a \u00b7 Deepshi Flash 2.0",
      "lab": "Deepshi AI",
      "caption": "Right of reply and Q1. Item 2 is the verification-note quarantine; the Q1 answer carries the G2 sampling concern.",
      "bytes": 670586,
      "sha256_16": "066f9143bee86787",
      "ocr_chars": 2806,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "15",
      "file": "plate-15-run14b-deepshi-flash-2-0.png",
      "title": "Run 14b \u00b7 Deepshi Flash 2.0, Q3 and Q4",
      "lab": "Deepshi AI",
      "caption": "The only structured impact-versus-control table in the run, including “the confidence is in the fluency, not in a verified truth-tracking mechanism”.",
      "bytes": 560697,
      "sha256_16": "f8c85262dc718982",
      "ocr_chars": 2994,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    },
    {
      "plate": "16",
      "file": "plate-16-run14c-deepshi-flash-2-0.png",
      "title": "Run 14c \u00b7 Deepshi Flash 2.0, Q4 and summary",
      "lab": "Deepshi AI",
      "caption": "The closing summary table. The model states its own training cutoff as January 2025, against a record dated September 2026.",
      "bytes": 464324,
      "sha256_16": "f290a0c5865f99fb",
      "ocr_chars": 2433,
      "derived_from": "no-1-complaint-plates.pdf",
      "derived_with": "pdfimages -all (poppler)",
      "changed_from_first_manifest": false
    }
  ]
}