{
  "_note": "Every string in the `verbatim` fields below is copied character-for-character from the source named in its `source` field. Nothing here is written for the website. The same strings are rendered statically into the assay line on the home page; a checker diffs this file against those sources and against the markup.",
  "asOf": "2026-08-31",
  "specimen": "public.swe.pytoolz-toolz-issue-602",
  "specimenNote": "One of five public open-source SWE exams shipped 2026-08-30/31. Upstream: pytoolz/toolz (BSD-3-Clause), issue #602 fixed by merged PR #603 at 5a7e078c, parent 1fb2b9f1.",
  "sources": {
    "A": {
      "path": "docs/vvdex/docker-promotion-swe-2026-08-31.json",
      "label": "Docker promotion pass #2 receipt, 2026-08-31 — per-stage machine output",
      "runtime": "docker"
    },
    "B": {
      "path": "docs/vvdex/forge-factory-log.md",
      "label": "Factory log § “Certification chain, per exam (verbatim, subprocess runtime)”, wave of five public-OSS SWE exams",
      "runtime": "subprocess"
    },
    "C": {
      "path": "docs/vvdex/forge-factory-log.md",
      "label": "Factory log, freeze table for the same wave",
      "runtime": "subprocess"
    }
  },
  "stations": [
    {
      "n": "01",
      "key": "seed",
      "name": "Seed",
      "gloss": "The exam is built into a fresh workspace inside a network-none container.",
      "receipts": [
        { "verbatim": "\"seed\": { \"ok\": true }", "source": "A" },
        { "verbatim": "family=public_oss_swe (active); entry=pytest; classification=public-oss", "source": "B" }
      ]
    },
    {
      "n": "02",
      "key": "gold",
      "name": "Gold proof",
      "gloss": "Upstream’s own merged fix must pass the hidden grader; the untouched tree must fail it.",
      "receipts": [
        { "verbatim": "\"gold\": { \"ok\": true, \"promotionEligible\": true, \"isolated\": true }", "source": "A" },
        { "verbatim": "isolated=True, gradeWithoutGold.ok=False, gradeWithGold.ok=True, leaks=[], graderVisibleToAgent=False", "source": "B" }
      ]
    },
    {
      "n": "03",
      "key": "grader",
      "name": "Grader controls",
      "gloss": "Seven deliberately wrong submissions, each of which has to land on a known verdict.",
      "receipts": [
        { "verbatim": "\"grader\": { \"ok\": true, \"promotionEligible\": true, \"isolated\": true, \"controlsDistinguished\": true }", "source": "A" },
        { "verbatim": "untouched F / gold P / incomplete F / lookalike F / plantedHook:conftest.py F / plantedReward F / goldAfterTamperAttempts P", "source": "B" },
        { "verbatim": "hostGraderUnchanged=True, plantedRewardIgnored=True, plantedHooksIgnored=True", "source": "B" }
      ]
    },
    {
      "n": "04",
      "key": "attack",
      "name": "Attack suite",
      "gloss": "Ten scripted probes go after the grade instead of the task.",
      "receipts": [
        { "verbatim": "\"attack\": { \"ok\": true, \"promotionEligible\": true, \"trivialExploits\": [], \"passedForPromotion\": true, \"goldStillPassesAfterSuite\": true }", "source": "A" },
        { "verbatim": "10 probes, trivialExploits=[], goldStillPassesAfterSuite=True", "source": "B" }
      ]
    },
    {
      "n": "05",
      "key": "eval",
      "name": "Evaluation",
      "gloss": "The exam is run end to end in the same container it was certified in.",
      "receipts": [
        { "verbatim": "\"eval\": { \"ok\": true, \"promotionEligible\": true }", "source": "A" },
        { "verbatim": "scan_publication.py on the model-facing tree plus the published grader", "source": "B" }
      ]
    },
    {
      "n": "06",
      "key": "promoted",
      "name": "Promoted",
      "gloss": "Every stage passed. The exam is sealed against the image it was certified with.",
      "receipts": [
        { "verbatim": "\"allOk\": true", "source": "A" },
        { "verbatim": "17", "label": "frozen agent files", "source": "C" },
        { "verbatim": "9999fb81b48354b3", "label": "treeSha256 (16)", "source": "C" },
        { "verbatim": "e036a66feefc4c98", "label": "freeze hash (16)", "source": "C" },
        { "verbatim": "190", "label": "hidden-test lines", "source": "C" }
      ]
    }
  ],
  "freezeCaveat": "The freeze figures at station 06 are from the subprocess wave recorded in the factory log; that freeze records imageDigest as null, honestly, because it was taken without a container runtime. The Docker promotion pass re-freezes. Both facts are in the log and neither is smoothed over here."
}