Factory docs, home

A record from the Factory repository, docs/intelligence-priority-receipt.json, shown as committed except that its local paths appear as placeholders in angle brackets, such as <local evidence archive>, standing for files that are not published; a placeholder that stands in for a key keeps the file's name, as in <local evidence file: report.md>. Every hash, date and count is as recorded. The build review says what the Factory's receipts support.

{
  "recordedAt": "2026-10-01T19:30:00Z",
  "component": "Intelligence (src/intelligence.ts): provider-neutral routes and per-task priority lists",
  "base": "d73b06e",
  "branch": "factory/intelligence-priority",
  "disposition": "merged-under-owner-gate: Opus 5.5 primary review PASS and Fable 5.1 (xhigh) secondary review PASS at 5a96159, coordinator judgement, 1 October 2026; no Astra acceptance claimed",
  "scope": "src/intelligence.ts only, with its tests, docs and the packaged adoption guide. Routes become declared data naming their provider; a task may name up to three `then` entries after its `pin`; the selector walks the list once, before any provider contact, records each pass-over, and never chooses outside the list; a checker never shares a model with the task it checks; the Claude facts record the owner's decision (a); discovery's triage and weekly synthesis name a proposed OpenAI entry first. Nothing is qualified, nothing dispatches, no live model call was made and nothing was pushed.",
  "plan": {
    "path": "<local evidence archive>",
    "revision": 2,
    "feedback": "<local evidence archive>"
  },
  "producer": "Sonnet 5.5 (claude-sonnet-5-5) at high effort implemented from the plan; it is the implementer and not the verifier. This follows the owner's 29 September 2026 direction (Opus plans, Sonnet or Sol implements), a departure from AGENTS.md rule 3's \"Opus 5.5 implements\". Astra (gpt-6-astra) reviewed round 1; the merge gate is recorded under mergeGate.",
  "ownerDecisions": {
    "a": "1 October 2026, coordinator session: keep the Claude CLI unavailable for the Factory's model roles (no Factory Claude profile; audit trust assumptions O2-O7 not accepted). Recorded as `unavailable` on every Claude fact; the coordinator's own Claude sessions are not what it rules out.",
    "b": "1 October 2026: discovery's daily and weekly triage and weekly synthesis switch to Codex, and the intelligence module gets a way to provide models at a priority for certain tasks so that arbitrary providers can be put in. Implemented as proposals (every OpenAI entry proposed until lifecycle L7).",
    "c": "1 October 2026, owner statement only: paid extra usage on the owner's ChatGPT account is off. The formal dated owner-ledger record belongs to roadmap item L-S1 and is not made here.",
    "e": "The Codex CLI source download (openai/codex rust-v0.159.1, commit dd48cd3d094a133f2781ab10343dc665a37e5bc0, tarball sha256:656cafb9f8bcbb247159709aa73df5a25ccdfd8d0e3c94e72c8fc5cf5b36227b) was approved and done by the owner; nothing was built or run for this change."
  },
  "localEvidence": {
    "note": "No live model call. The OpenAI model facts rest on the local Codex model catalogue only, as read once by the implementer; a catalogue listing is not entitlement. The cache is rewritten by whichever installed Codex client runs, without notice, so only this recorded hash and extract are fixed.",
    "modelsCache": {
      "path": "<home directory path>",
      "readAt": "2026-10-01T18:12:01.271Z",
      "bytes": 443424,
      "sha256": "sha256:e70cf9a43676989943683790c0240e8302e0e8a725d0aec1373b63305bd561a6",
      "etagSha256": "sha256:7074c5efd40b18a32d0f8eee3a21308a28ddec61f5ea830fad3fad4c4dc74bf9",
      "etagNote": "SHA-256 of the cache's exact etag string; the etag itself is a 32-hex value the public build refuses as a possible private identifier, and it stays in the archive extract and the plan's Appendix A",
      "fetched_at": "2026-10-01T18:10:58.045939Z",
      "client_version": "0.159.3",
      "extract": "<local evidence archive>",
      "excluded": "the `identity` field was never read or copied; auth.json was never opened",
      "planner": "<local evidence archive> read at 17:30:27Z (client 0.159.1, sha256:c81da4c7e4b387534a8172581cb7349844ccd02210a701c5d86d99abd25bfb7a); the three models' catalogue facts are identical across both reads"
    }
  },
  "behaviourChange": "Provider and Transport widen from closed unions to grammar-checked strings; routes are declared data (`ModelTiers.transports`, `TransportDeclaration`, frozen `TRANSPORTS_V1` default, `INTELLIGENCE_LIMITS` {entries: 4, transports: 16}); `openai-responses` is no route. `IntelligenceTask` gains `then` (one to three entries after `pin`) and `checks`; `ModelPin` gains `proposed: true`. `ModelRequest.requested` becomes a union of the choose form (policy only) and the verify form (an entry named in full); `escalation` needs the verify form; `ModelChoice` gains `entry` and `skipped`; `ModelSelection` gains `no-eligible-entry` carrying `skipped`; `RefusalCode` gains `proposed` and `no-eligible-entry`. Requests, producer identities and failed escalations name providers and routes by grammar only: a well-formed undeclared name was a TypeError and is now a refusal value (`pin-mismatch`, `producer-unknown`, `escalation-mismatch`). New load rules for routes, lists, proposed entries and `checks` independence. Every Claude fact is `unavailable` (owner decision (a)) and asOf is 2026-10-01; facts gpt-6-luna and gpt-6.1-sol are listed on codex-cli (unqualified, no effort, request count unknown); triage and synthesis name a proposed OpenAI entry first. One-entry tasks behave, refuse and digest as before given the same facts.",
  "digestEncoding": "`sf-intelligence/1`, extended compatibly: `then` enters the canonical form only where declared and `proposed: true` only on a proposed entry; no other key is added. A one-entry task encodes byte-identically (stage A reproduces all eleven recorded digests; tests also pin a one-entry literal computed from the source before the change).",
  "policyDigests": {
    "method": "taskPolicyDigest for every task, from the real code at each stage (snapshot.mjs in the archive; the planner's digest-prototype reproduces the same values)",
    "before": {
      "project-guidance/summarise": "sha256:d80572acf57a2d8295820c18099f4b96dea310dbf4cdfda20edff0ec887afd2a",
      "producer/propose": "sha256:c968ae728c657db94b3de4a07c86be013d3a9462aa3c503be5280a8011909375",
      "repository-assurance/review": "sha256:576b016aaac6d4d0e4fb5cd381c95c530edb1cf89d435969a23bc02a232ac963",
      "repository-assurance/fresh-scenario": "sha256:e0d63054ba0fdb23647381112cafd457df8e38e0b42e56506360febfb44b3567",
      "brief/explain-trade-offs": "sha256:9ab750bbfab79dfc494d2cb557e4c78cd991400a03becfe21602f0220ac1702a",
      "planning/propose-improvement": "sha256:cc1e170b76dec4cdb34121439343625f6500590854c7448d58ee3eefd79c52ba",
      "signal-intake/triage": "sha256:39e85bcdbafe046bccd6ee7834fd120d58dc71a050b921381d9d332c5d09f5b1",
      "signal-intake/search": "sha256:97d3de4cf937115d9bb26675e99de97a73f79e1b41ade55cdc1bc47815c0d5bb",
      "opportunity-synthesis/propose": "sha256:9520c8c24e984ab93e4a2935bbc722a4e68cef7d6b4616b0039e52e8871babe6",
      "growth/draft-pack": "sha256:cb99fbe11548af06a025b97ff84146db4ef32291ffc6388e34bc5894baa777ae",
      "capability-retirement/draft-case": "sha256:0204c52414a5aafbfcfe83cbae19d7674f4e07eea29ca3f7ec3d34fcf3effb20"
    },
    "stageA": {
      "description": "mechanism only, data unchanged: byte-identical to before for all eleven tasks",
      "digests": {
        "project-guidance/summarise": "sha256:d80572acf57a2d8295820c18099f4b96dea310dbf4cdfda20edff0ec887afd2a",
        "producer/propose": "sha256:c968ae728c657db94b3de4a07c86be013d3a9462aa3c503be5280a8011909375",
        "repository-assurance/review": "sha256:576b016aaac6d4d0e4fb5cd381c95c530edb1cf89d435969a23bc02a232ac963",
        "repository-assurance/fresh-scenario": "sha256:e0d63054ba0fdb23647381112cafd457df8e38e0b42e56506360febfb44b3567",
        "brief/explain-trade-offs": "sha256:9ab750bbfab79dfc494d2cb557e4c78cd991400a03becfe21602f0220ac1702a",
        "planning/propose-improvement": "sha256:cc1e170b76dec4cdb34121439343625f6500590854c7448d58ee3eefd79c52ba",
        "signal-intake/triage": "sha256:39e85bcdbafe046bccd6ee7834fd120d58dc71a050b921381d9d332c5d09f5b1",
        "signal-intake/search": "sha256:97d3de4cf937115d9bb26675e99de97a73f79e1b41ade55cdc1bc47815c0d5bb",
        "opportunity-synthesis/propose": "sha256:9520c8c24e984ab93e4a2935bbc722a4e68cef7d6b4616b0039e52e8871babe6",
        "growth/draft-pack": "sha256:cb99fbe11548af06a025b97ff84146db4ef32291ffc6388e34bc5894baa777ae",
        "capability-retirement/draft-case": "sha256:0204c52414a5aafbfcfe83cbae19d7674f4e07eea29ca3f7ec3d34fcf3effb20"
      },
      "identicalToBefore": true
    },
    "stageB": {
      "description": "owner decision (a): the five Claude facts `unavailable`, asOf 2026-10-01; signal-intake/search unchanged because its research fact was already `unavailable`",
      "digests": {
        "project-guidance/summarise": "sha256:7c1fb0bf81b3fbc9cfa57c82ca08d4a0520f00486e25c34158c4563df0a2610a",
        "producer/propose": "sha256:49fd99e728748541130c32155dab0f2785de1c10eea913869af47dc81f1ed4cf",
        "repository-assurance/review": "sha256:84fa441d9e57c862d5c156a31bee888d6d0ffd78d628fce6e83279cca11e935a",
        "repository-assurance/fresh-scenario": "sha256:cbb341c0e823cf2652ddd2107650229b9f3d1ef817ab9244641340606db84269",
        "brief/explain-trade-offs": "sha256:f29fee3fb769e68d5a15b8e9d39457029f2fe12fad2298f7a4408cdfec0d6100",
        "planning/propose-improvement": "sha256:28680da2abf0614530b814c3105129c20d95173598f2b0d3168f86110e7c2655",
        "signal-intake/triage": "sha256:ad6a381450de6a83a93e990ed332dc39503cf96567656f1969a210f5650d27b0",
        "signal-intake/search": "sha256:97d3de4cf937115d9bb26675e99de97a73f79e1b41ade55cdc1bc47815c0d5bb",
        "opportunity-synthesis/propose": "sha256:43b8e0b0923466bdf7bfe069800e9e580a44831227ace66709f3a53d6d4d42c7",
        "growth/draft-pack": "sha256:e6957386edecff5b794bbe94a9b3a485366a2fbc93807edeb082b4b8ee5fa831",
        "capability-retirement/draft-case": "sha256:f6007d2fdaf5a779d14547bd07fd06bf01c20ce3a6bc444eba330bc7dd329c48"
      }
    },
    "after": {
      "project-guidance/summarise": "sha256:7c1fb0bf81b3fbc9cfa57c82ca08d4a0520f00486e25c34158c4563df0a2610a",
      "producer/propose": "sha256:49fd99e728748541130c32155dab0f2785de1c10eea913869af47dc81f1ed4cf",
      "repository-assurance/review": "sha256:84fa441d9e57c862d5c156a31bee888d6d0ffd78d628fce6e83279cca11e935a",
      "repository-assurance/fresh-scenario": "sha256:cbb341c0e823cf2652ddd2107650229b9f3d1ef817ab9244641340606db84269",
      "brief/explain-trade-offs": "sha256:f29fee3fb769e68d5a15b8e9d39457029f2fe12fad2298f7a4408cdfec0d6100",
      "planning/propose-improvement": "sha256:28680da2abf0614530b814c3105129c20d95173598f2b0d3168f86110e7c2655",
      "signal-intake/triage": "sha256:8d986678bedcb142e1df1e55504d1d6ae49bbd3dc07660baec3dc9ac586602ca",
      "signal-intake/search": "sha256:97d3de4cf937115d9bb26675e99de97a73f79e1b41ade55cdc1bc47815c0d5bb",
      "opportunity-synthesis/propose": "sha256:35298c75d7c425cd18ed86c1fc6c2dda54de7bc38a623d030ec6533099f7f035",
      "growth/draft-pack": "sha256:e6957386edecff5b794bbe94a9b3a485366a2fbc93807edeb082b4b8ee5fa831",
      "capability-retirement/draft-case": "sha256:f6007d2fdaf5a779d14547bd07fd06bf01c20ce3a6bc444eba330bc7dd329c48"
    },
    "changed": {
      "project-guidance/summarise": {
        "before": "sha256:d80572acf57a2d8295820c18099f4b96dea310dbf4cdfda20edff0ec887afd2a",
        "afterStageB": "sha256:7c1fb0bf81b3fbc9cfa57c82ca08d4a0520f00486e25c34158c4563df0a2610a",
        "final": "sha256:7c1fb0bf81b3fbc9cfa57c82ca08d4a0520f00486e25c34158c4563df0a2610a",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "producer/propose": {
        "before": "sha256:c968ae728c657db94b3de4a07c86be013d3a9462aa3c503be5280a8011909375",
        "afterStageB": "sha256:49fd99e728748541130c32155dab0f2785de1c10eea913869af47dc81f1ed4cf",
        "final": "sha256:49fd99e728748541130c32155dab0f2785de1c10eea913869af47dc81f1ed4cf",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "repository-assurance/review": {
        "before": "sha256:576b016aaac6d4d0e4fb5cd381c95c530edb1cf89d435969a23bc02a232ac963",
        "afterStageB": "sha256:84fa441d9e57c862d5c156a31bee888d6d0ffd78d628fce6e83279cca11e935a",
        "final": "sha256:84fa441d9e57c862d5c156a31bee888d6d0ffd78d628fce6e83279cca11e935a",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "repository-assurance/fresh-scenario": {
        "before": "sha256:e0d63054ba0fdb23647381112cafd457df8e38e0b42e56506360febfb44b3567",
        "afterStageB": "sha256:cbb341c0e823cf2652ddd2107650229b9f3d1ef817ab9244641340606db84269",
        "final": "sha256:cbb341c0e823cf2652ddd2107650229b9f3d1ef817ab9244641340606db84269",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "brief/explain-trade-offs": {
        "before": "sha256:9ab750bbfab79dfc494d2cb557e4c78cd991400a03becfe21602f0220ac1702a",
        "afterStageB": "sha256:f29fee3fb769e68d5a15b8e9d39457029f2fe12fad2298f7a4408cdfec0d6100",
        "final": "sha256:f29fee3fb769e68d5a15b8e9d39457029f2fe12fad2298f7a4408cdfec0d6100",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "planning/propose-improvement": {
        "before": "sha256:cc1e170b76dec4cdb34121439343625f6500590854c7448d58ee3eefd79c52ba",
        "afterStageB": "sha256:28680da2abf0614530b814c3105129c20d95173598f2b0d3168f86110e7c2655",
        "final": "sha256:28680da2abf0614530b814c3105129c20d95173598f2b0d3168f86110e7c2655",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "signal-intake/triage": {
        "before": "sha256:39e85bcdbafe046bccd6ee7834fd120d58dc71a050b921381d9d332c5d09f5b1",
        "afterStageB": "sha256:ad6a381450de6a83a93e990ed332dc39503cf96567656f1969a210f5650d27b0",
        "final": "sha256:8d986678bedcb142e1df1e55504d1d6ae49bbd3dc07660baec3dc9ac586602ca",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field",
          "owner request (b): its `pin` is now an OpenAI entry through codex-cli, proposed, and its earlier Claude default moved to `then`"
        ]
      },
      "opportunity-synthesis/propose": {
        "before": "sha256:9520c8c24e984ab93e4a2935bbc722a4e68cef7d6b4616b0039e52e8871babe6",
        "afterStageB": "sha256:43b8e0b0923466bdf7bfe069800e9e580a44831227ace66709f3a53d6d4d42c7",
        "final": "sha256:35298c75d7c425cd18ed86c1fc6c2dda54de7bc38a623d030ec6533099f7f035",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field",
          "owner request (b): its `pin` is now an OpenAI entry through codex-cli, proposed, and its earlier Claude default moved to `then`"
        ]
      },
      "growth/draft-pack": {
        "before": "sha256:cb99fbe11548af06a025b97ff84146db4ef32291ffc6388e34bc5894baa777ae",
        "afterStageB": "sha256:e6957386edecff5b794bbe94a9b3a485366a2fbc93807edeb082b4b8ee5fa831",
        "final": "sha256:e6957386edecff5b794bbe94a9b3a485366a2fbc93807edeb082b4b8ee5fa831",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      },
      "capability-retirement/draft-case": {
        "before": "sha256:0204c52414a5aafbfcfe83cbae19d7674f4e07eea29ca3f7ec3d34fcf3effb20",
        "afterStageB": "sha256:f6007d2fdaf5a779d14547bd07fd06bf01c20ce3a6bc444eba330bc7dd329c48",
        "final": "sha256:f6007d2fdaf5a779d14547bd07fd06bf01c20ce3a6bc444eba330bc7dd329c48",
        "why": [
          "its listed Claude fact (claude-cli, or Fable 5.1 for project-guidance/summarise) is now `unavailable` by the owner's decision (a) of 1 October 2026; the qualification state is an enforced field"
        ]
      }
    },
    "unchanged": [
      "signal-intake/search"
    ],
    "blockedWithoutPolicy": [
      "planning/challenge-opportunity",
      "signal-intake/audit",
      "growth/review-pack"
    ],
    "silentRepins": "none: no frozen request or stored task policy digest exists anywhere (only src/intelligence.ts names chooseModel and taskPolicyDigest). Historical receipts are byte-unchanged; test/intelligence.test.ts reproduces the eleven historical digests from an in-file reconstruction of the earlier registry and pins the eleven final digests as literals."
  },
  "files": {
    "README.md": "6671684abfc5dcdf9fa701baf36cb9053abc75e7cb4167d7dbdf022f118f1d12",
    "docs/adoption-inventory.md": "c2c2f8f44291c7cf211b56474580675853d850b9c1896ca9d5a7f619ac7d6e1e",
    "docs/agents/factory-model.md": "85ca93db9d6f7825ee573183e9ebec8270c88b3fc15c7e7578620f276a57f151",
    "docs/agents/intelligence.md": "7b2cc160f8cdaa0a2d408ae231b8fb2da676b1c3fd079df89abddf9d0edf3e37",
    "docs/agents/observability.md": "abf658e6e5c9e3669b20f923d1e3f8a0a95ea9dad11b2820cfa9c782deb0c02b",
    "docs/build-review.md": "a5b31234fc9a7c3e4ba8189237d9a323cdffccbcc956515e714429189a7a5c1d",
    "docs/guide/intelligence.md": "ace34565aba97e73e25f6776281b713a48f3805a98be641737841d82e2af1848",
    "docs/how-to/adopt-foundations.md": "61d215b23ff9f1cd243a02c507f10f6fae7986ed72f621f7a8d3cf957c821bfb",
    "docs/lifecycle.md": "840b2572541112afbe943ee58138c60bbd94167ca14378781410ed4a048456c2",
    "site/agents/factory-model.html": "b193f36e5896b5e2bdc88ccd9de67d6092441c5c372d3de535a787d03c550910",
    "site/agents/intelligence.html": "a88a612f996ef0babdf3d259701f3aa9a8b2e36a5fe143811235fbbcd0817d0a",
    "site/agents/observability.html": "a8abdae62f46ba0e77772c667c563af99cb849127d85553fa08f30b8e14c3d88",
    "site/build-manifest.json": "8d7b3605d2a05a1471a0179d9bfbee4e16f78117c4d2f6e68f158fdb214f8931",
    "site/intelligence.html": "f6692947e94b3b95c01bccf160fe76e54381b5bc705ae1cab37cc58197fd10c4",
    "site/reference/adoption-inventory.html": "836c0348f4380fa5026a506141d9092566353d0c929f24189fdc3fe3df3a6ec5",
    "site/reference/build-review.html": "eccb5975a9fefa77489ed7f830e87030b67fc3ce1d70185efafe55bd1d55d89f",
    "site/reference/how-to/adopt-foundations.html": "b501b51f5897e68bf12ca13ff125c2dad42e946398b21cc794e14cc743192174",
    "site/reference/lifecycle.html": "664bd08888aa1d486dcfb71c581a529f15a318032a7938b0ca8fdf4ed0e41859",
    "src/intelligence.ts": "b9da1093c99839b582ac8e7f910ca3bb834684d3d500907aa9dadb5a02d96610",
    "test/distribution.test.ts": "521740cefcbec732dc843859a74c9a5ab2bb818649aa2f143e7b1aada2a9c88a",
    "test/fixtures/distribution/foundations.mjs": "dfff8d536eefe66b927b42fd4efae4b549dbed6ac0e30c5ce770437d1348abb9",
    "test/intelligence-lifecycle-adversarial.test.ts": "4e350ceefe9249c59c10ef908d085b8f54da79081887c80b4dc86a00f322e799",
    "test/intelligence-lifecycle.test.ts": "e9781f8bfbb245bb29363338f5be271cb90807a308f28b4c512919dbaebcc60b",
    "test/intelligence-priority.test.ts": "639b4daac6b68b814e7eeb8793a6ec8fa31c8b51f3c41c6545f5f5a3d30a047f",
    "test/intelligence.test.ts": "3f4f141de6fa1b02ce5700894c05d59dcb0bd17c0136df62485cce35b005bae2"
  },
  "testChanges": {
    "new": "test/intelligence-priority.test.ts (17 tests, fixtures only): routes as data; undeclared or malformed routes refused at load; order respected; pass-overs recorded with exact reasons; exhaustive walk (4^4 drafter and 5^4 verifier combinations, choose and verify forms agreeing); none eligible; unknown is never zero; malformed lists refused; the two forms; F3; run-time and load-time independence; the digest covers the whole list; stable digests for unchanged tasks (with a literal computed from the source before the change); reproducibility; purity; the reopened Claude route.",
    "updated": [
      "test/intelligence.test.ts: the exact producer choice gains `entry: 1, skipped: []` (an added expectation, nothing removed); the per-task pin assertion applies to every established entry of the list and proposed entries get their own assertions; a new assertion holds that every Claude fact is `unavailable` with the owner's decision first; the eleven-literal digest test now reproduces the eleven historical digests from an in-file reconstruction of the earlier registry and pins the eleven final digests as literals. Unchanged: the registry-check cases and the request cases that name `openai-responses`.",
      "test/intelligence-lifecycle.test.ts: triage and synthesis expectations become their two-entry lists (the Claude entry's earlier assertions apply to the last entry); 'no task pins either OpenAI route' becomes 'every OpenAI entry is proposed, no established entry is on codex-cli, nothing is listed on openai-responses'; the efficient-why assertion becomes a table of the comparison each efficient task's first entry waits on (triage names the gpt-6.1-sol low baseline the owner has not chosen); the 'no model becomes selectable' test expects `proposed` where the first entry is proposed and adds the choose form (`no-eligible-entry` with codes proposed, not-qualified).",
      "test/intelligence-lifecycle-adversarial.test.ts: the two line anchors (src/intelligence.ts:972 and docs/agents/intelligence.md:94), and its helper now qualifies the first established entry of triage's list (the proposed gpt-6-luna entry is skipped), so tests 1 and 2 reach their gates again (draft-pack and search refuse not-measured; search, once measured, not-qualified); nothing removed or loosened.",
      "test/fixtures/distribution/foundations.mjs and test/distribution.test.ts: a new `intelligence-priority` mode and call under the read-only permission model, and typed-consumer lines for the new exports and request forms; the existing `intelligence` mode is unchanged."
    ],
    "weakened": "none: no assertion was removed or loosened; reasons are recorded per change above"
  },
  "adoptionGate": {
    "contract": "entry software-factory/intelligence, same name; exports and types per the reference; TypeError for malformed requests and registries, values for refusals, with the error-class change recorded; stable ids (task ids, refusal codes, sf-intelligence/1); bounds INTELLIGENCE_LIMITS and MAX_REASON_LENGTH; dependencies unchanged (node:crypto and canonical JSON); importing starts nothing.",
    "externalConsumer": "the packed tarball's `intelligence-priority` mode runs in a disposable application outside the repository under the read-only permission model (test/distribution.test.ts); the typed consumer typechecks the new exports and a @ts-expect-error for a partial naming group. Distribution diagnostic: artefact software-factory-0.1.0.tgz sha256 e663f4e3a5ba87a8c2df4fc4d95f846a66dfea10e7b5899d369eeac6723a5e4e, node v26.8.1, darwin arm64, passed in check-build.log.",
    "substitution": "a caller-supplied registry declaring an arbitrary route, provider and model selects through the same contract without consumer edits; it supplies policy data only (no qualification: closed Producers ProviderId; no admission: closed schedule Account; no authority)",
    "failureAndState": "malformed routes, over-long lists, proposed-entry misuse and independence violations are refused at load; stale policies are refused; the module keeps no state, so there is no migration; every changed digest is enumerated under policyDigests",
    "composition": "the Factory uses the same contract (no src/ file imports it, unchanged); the packaged closure stays exactly eight files; admission of a task with `then` through the accepted schedule is not established (FU-1)",
    "closed": "no adoption gate and no epic is closed by this; Fable acceptance stays pending"
  },
  "admissionPrerequisite": "NOT ESTABLISHED. The accepted schedule maps tasks to accounts statically (`proposedCommission` names `claude` for triage and synthesis) and `admitCall` takes and records no model, so a task with `then` is not admissible through it until FU-1 lands.",
  "verification": {
    "typecheck": "npx tsc -p tsconfig.json: clean",
    "focused": "node --test test/intelligence*.test.ts test/module-graph.test.ts test/docs-site.test.ts test/design-tokens.test.ts test/documentation.test.ts test/style-lint.test.ts test/readability.test.ts test/website.test.ts: 199 tests, 199 pass, 0 fail (focused-tests.log)",
    "stageGates": "stage A: all eleven digests byte-identical to the recorded ones; stage B: exactly the ten digests of the plan's table (search unchanged) and no refusal code changes for any task or request, today or with protocols specified and efficient tasks measured (stageb-codes-branch.txt: differences 0); stage C: only triage and synthesis change from stage B, matching the plan's prototype values",
    "fullCheck": "npm run check: 2149 tests, 2145 pass, 4 fail, 0 skipped (check-build.log). The four failures were environmental, not of this change, and each file passes alone on the final tree (rerun-failed-files-4.log to -6.log): test/docs-site.test.ts and test/repository.test.ts (with the repository's own tmp directory and TMPDIR conventions: the first run used a symlinked tmp to dodge a full internal disk, which the docs-site 'inside the repository' check rightly refuses; repository.test.ts asserts a /private/var temporary directory, which the Memory TMPDIR is not) and test/provider-guidance-tools.test.ts and test/provider-subscriptions-symbol.test.ts (their 60 s and 30 s child timeouts under a machine load average above 40; five of five pass alone). An earlier full run (check-build-run2-env-failures.log: 2149 tests, 24 failures) also had ENOSPC failures in test/discovery-sources*.test.ts and one genuine failure, test/observability-catalogue.test.ts, which listed RefusalCode without the two new codes: fixed in docs/agents/observability.md, and it passes. An even earlier aborted run is check-build-aborted-run1.log. The distribution test passed in the full run, including the new intelligence-priority mode.",
    "environment": "TMPDIR was set to <local cache> for every test run, because the internal disk was at 99% (165 MB free) and an earlier distribution run failed with ENOSPC; node v26.8.1 on macOS"
  },
  "evidenceSha256": {
    "<local evidence archive: check-build.log>": "01e88886cac93c678ebab2ab600d53ce7cc60b3cb3fa5a5f45f6bebcc76b2f17",
    "<local evidence archive: distribution-run1.log>": "c9e559863e7790ab56956d84e77d8b21745db35f99997c4a0cc8514222d4a0ee",
    "<local evidence archive: focused-tests.log>": "62474dee17c673c563ebcabf11630bc84782bf1c194ff4fa4d1f7c08c436d481",
    "<local evidence archive: policy-digests-before.json>": "ab97e6827b0c4be55bbf3ea5914a6bb159531c3bf777ec0009e897272d4130ea",
    "<local evidence archive: policy-digests-stageA.json>": "ab97e6827b0c4be55bbf3ea5914a6bb159531c3bf777ec0009e897272d4130ea",
    "<local evidence archive: policy-digests-stageB.json>": "5bce5eb1b758aaa85b048c55dd5b9d1f001368683149796a29d9b47396caa483",
    "<local evidence archive: policy-digests-after.json>": "e073468a92a2c6519545781d5a13b555be3a3df0af6f8cbb2fec209564eb10ad",
    "<local evidence archive: models-cache-extract-implementer-20261001.json>": "87c78632e8ccf71c0f4c990a1ebad5bcfabd90b4ef240b8d694a172e69e89c41",
    "<local evidence archive: stageb-codes-branch.txt>": "69b31fc556a8d5f5c7679caa882f09c694f273cf4b15a5acb930c835431746ca",
    "<local evidence archive: check-coord.log>": "4cba3521ec057b69b52b8c4a7136e5fe0786568a347dfbf6a1c3206a72b646d8",
    "<local evidence archive: check-coord-run1-etag-and-load-failures.log>": "93bf289cd2e303e69397931a8b87c74999b3bdc7c741a0cd0d5a8ea7a850c85f",
    "<local evidence archive: coord-focused-2.log>": "d607dbf6280d387a4c73fb9d53f5e18c1a35042ac4696c9629c87db6db438a8a",
    "<local evidence archive: coord-rerun-distribution.log>": "9a7a7ecbac355e554f4b88a3e60a32cde533b85225a1d6db19f20784954d6db6",
    "<local evidence archive: coord-rerun-rpc-x3.log>": "11acea46b9ed806e8baac3c3b82b0f73c12ec28e7a6b482e07194f5ffee98736",
    "<local evidence archive: coord-rerun-public-build.log>": "06cfedf5888f22225677df84ec56687368708c73b9372573337b4ba02bac8289",
    "<local evidence archive: check-r1.log>": "8866a8c128fbc0bf992dd226bfe506a3a31422a67060cfeabdf949f1e06b78cf",
    "<local evidence archive: check-r1-rerun-repository.log>": "913524d418265c66ff701333e1167fa13dbabc60655a6c21f94ac491689c06fd",
    "<local evidence archive: astra-r1.md>": "129c42d860ea97a8f5f3b3e890e0def799a6b078993ff1a5abd10bd0c7abdc84",
    "<local evidence archive: fable-r1.md>": "eab1b4f07285a3c55b9636385379e42736f2c81bcc4a68883646e41e10537e38",
    "<local evidence archive: dispositions-r1.md>": "a981b838d38907e8812007b881af8e6f6c2904447a4cf3b712c678e9773724d4"
  },
  "reviews": [
    {
      "report": "<local evidence archive>",
      "reviewer": "Astra (gpt-6-astra)",
      "round": 1,
      "verdict": "FAIL",
      "findings": "a stray `proposed` key on an ordinary task bypassed the static independence rule; three independence tests never reached their rule"
    },
    {
      "report": "<local evidence archive>",
      "reviewer": "Fable 5.1",
      "round": 1,
      "verdict": "FAIL",
      "findings": "the same two defects (F1, F2); no other defect; notes only"
    },
    {
      "report": "<local evidence archive>",
      "reviewer": "Opus 5.5 (claude-opus-5-5), xhigh, primary merge review; not the implementer",
      "round": "m1",
      "verdict": "PASS",
      "findings": "no material defect; notes N1-N7 (N1 vacuous lifecycle-adversarial tests 1-2, N2 check-order sentence, N3 rule 3 departure, N4 merge bookkeeping applied in the final commit; N5-N7 left as follow-ups)"
    },
    {
      "report": "<local evidence archive>",
      "reviewer": "Fable 5.1 (claude-fable-5-1), xhigh, independent secondary review",
      "round": "m1",
      "verdict": "PASS",
      "findings": "no material defect; notes on merge bookkeeping, reference wording (EntryRefusal/EntrySkip names, check order) applied or left as follow-ups"
    }
  ],
  "followUps": {
    "FU-1": "prerequisite before any task with `then` is admitted: lifecycle schedule `admitCall` records the choose form's result (provider, model, effort, route, entry) with the task's policy digest and derives the account from the chosen entry's provider; Account stays closed until a new subscription is added by its own review",
    "FU-2": "L7 scope grows: audit and transport for gpt-6-luna and gpt-6.1-sol as well as gpt-6-astra; establish codex-cli's input handling and limit and its efforts; then the proposed entries and the three blocked tasks become established by reviewed edits, with a record of the checked dispatch's model for drafter-checkers or the gap accepted explicitly",
    "FU-3": "roadmap: L6, FQ and L10 are blocked by decision (a); the C1-* audit lanes lose their purpose for the Factory's roles (for the roadmap lane, not this change)",
    "FU-4": "MRG0 (symbol anchors) would end the line-anchor edits this change needs",
    "FU-5": "website copy 'which one, at what effort' to 'which models, in what order' (website lane)",
    "OD1-OD9": "owner decisions with defaults, per plan section 15: rule 7 wording, F3 clarification, weekly search on Codex, L3 baseline for OpenAI efficient entries (gpt-6.1-sol low proposed), same-provider checking, Codex entries for the other Claude-only tasks, the proposed models and their order, whether decision (a) is represented now, account-aware selection"
  },
  "notEstablished": [
    "any qualified model; a codex-cli Producers transport, audit, commission or request count (L7); that codex exec runs headless with --output-schema on this sign-in for gpt-6-luna, gpt-6.1-sol or gpt-6-astra; how the Factory would set an effort on codex-cli; that Codex refuses rather than truncates or compacts an over-long input",
    "admission of a task with `then` through the accepted schedule (FU-1)",
    "arbitrary providers end to end: a caller-declared provider can be selected, but the Producers contract (`ProviderId`) cannot qualify it and the schedule (`Account`) cannot admit it",
    "that the cached catalogue is complete or current, or that every listed model is usable on this account's plan",
    "the quality or cost of the proposed models, that their efforts are right (each mirrors the former Claude default), L3 for any OpenAI entry, and the L3 baseline (OD4)",
    "independence beyond model identity; any record of a checked dispatch's actual model for drafter-checkers",
    "the L-S1 dated record for the ChatGPT account (owner statement (c) only)",
    "any adoption gate closed, any epic closed, any customer value; Fable acceptance stays pending; docs/foundations-adoption-draft.json's source hash for src/intelligence.ts and the earlier Intelligence receipts are now historical"
  ],
  "coordinatorReview": {
    "reviewer": "Opus 5.5 (claude-opus-5-5), coordinator reviewer; not the implementer and not the independent verifier",
    "date": "2026-10-01",
    "finding": "A proposed list entry's effort was checked by no rule: listedEntry checked route, model, date, marker, duplicates and tools, and a proposed entry is exempt from the established-entry effort and tier check, so a proposed entry at an effort outside EFFORTS (for example \"ultra\") loaded and entered the policy digest, while a blocked task's proposed pin already refused it. Fixed: every list entry's effort must be one of EFFORTS (src/intelligence.ts, listedEntry).",
    "testChange": "test/intelligence-priority.test.ts, the malformed-list test: a new case for a proposed entry at an effort outside EFFORTS; the case 'a proposed entry on an unlisted model' and the new case now also drop entry 2's tier candidate (helper unestablish), since emptying that fact's efforts alone made the registry fail on the stale tier candidate, so neither case tested its own rule; a control asserts the same mutation with a well-formed proposed entry loads. The new case fails on the unfixed source and passes on the fixed one. Nothing removed or weakened.",
    "docsChange": "docs/agents/intelligence.md invariant 2 names the effort rule; site/ rebuilt.",
    "policyDigests": "unchanged: no bundled entry changes",
    "filesRefreshed": "The implementer's recorded hashes for these five files were superseded by this review's edits; they remain in this file's Git history at the implementer's commit 9fd1080: docs/agents/intelligence.md, site/agents/intelligence.html, site/build-manifest.json, src/intelligence.ts, test/intelligence-priority.test.ts",
    "receiptChange": "localEvidence.modelsCache.etag replaced by etagSha256: the raw etag, a 32-hex string, made npm run publish:build (test/public-build.test.ts) refuse docs/intelligence-priority-receipt.json as carrying a private conversation identifier.",
    "verification": "npx tsc clean; focused node --test test/intelligence*.test.ts test/module-graph.test.ts test/observability-catalogue.test.ts: 85 pass, 0 fail (coord-focused-2.log). npm run check run 1: 2149 tests, 32 fail: 28 in test/public-build.test.ts and test/public-host.test.ts from the receipt's raw etag (fixed), and two load or timing failures in test/distribution.test.ts (dashboard consumer HTTP 503) and test/provider-events-rpc-consumer.test.ts, run straight after the internal disk had been full (check-coord-run1-etag-and-load-failures.log). Run 2 after the fix: 2149 tests, 2148 pass, 1 fail, test/provider-events-rpc-consumer.test.ts (0 !== 1 in its quiet-expiry fixture; it imports nothing of Intelligence), which passes alone three times out of three (coord-rerun-rpc-x3.log); test/distribution.test.ts alone 146 of 146 (coord-rerun-distribution.log); test/public-build.test.ts alone 21 of 21."
  },
  "round1Fixes": {
    "implementer": "Sonnet 5.5 (claude-sonnet-5-5), high; dispositions in <local evidence archive>: both findings confirmed by reproduction, none rejected",
    "srcChange": "src/intelligence.ts checkIndependence: the collection decides, never a key. Ordinary tasks list their entries and escalation; registry.blocked entries list their one proposed model; an ordinary task that carries a `proposed` key is refused after the shared-model rule. No line above the cited anchor src/intelligence.ts:972 moved; checks is outside the digest, so no policy digest changes.",
    "testChange": "test/intelligence-priority.test.ts, 'independence at load': an otherRoute helper keeps the checked task's pin valid; every case carries an expected-message regex; new cases for a leftover `proposed` key (fails on the unfixed source) and for the key alone; two loading controls. Nothing removed or weakened.",
    "docsChange": "docs/agents/intelligence.md invariant 9 (collection decides; stray key refused) and the 'Today every task refuses' wording (the weekly synthesis is frontier, so no measurement gate); docs/build-review.md row and section name round 1 FAIL and round 2 pending; site/ rebuilt.",
    "filesRefreshed": "Superseded hashes remain in this file's Git history at the previous commit: docs/agents/intelligence.md, docs/build-review.md, site/agents/intelligence.html, site/build-manifest.json, site/reference/build-review.html, src/intelligence.ts, test/intelligence-priority.test.ts",
    "verification": "npx tsc clean; test/intelligence-priority.test.ts 17/17; npm run check (check-r1.log): 2149 tests, 2148 pass, 1 fail, 0 skipped; the one failure was test/repository.test.ts, whose assertion expects a /private/var temporary directory and so fails under the TMPDIR on <local volume> used for the run (the same environmental failure as in the build run); it passes alone, 7/7, with the default TMPDIR (check-r1-rerun-repository.log). No timing test failed and none was rerun."
  },
  "mergeGate": {
    "ownerDirection": "1 October 2026, about 23:00 BST, to the coordinator: \"can we super prioritise getting the work merged, use fable on xhigh for secondary review and then when you're happy get it merged\"; then \"get them merged if they are ready\"",
    "gate": "Opus 5.5 primary review and Fable 5.1 (xhigh) secondary review, both PASS at 5a96159, then the coordinator's judgement and a full npm run check on the branch",
    "finalCommitChanges": "applies review notes N1 (test/intelligence-lifecycle-adversarial.test.ts helper qualifies the first established entry, so tests 1-2 reach their gates again), N2 (reference check-order sentence), N3 (rule 3 departure stated) and N4 (status lines, this disposition and reviews); site/ rebuilt; this receipt's file hashes refreshed, superseded hashes in Git history at 5a96159",
    "notClaimed": "Astra acceptance of the round 1 fixes; any qualification, transport, commission or dispatch"
  }
}

Source: docs/intelligence-priority-receipt.json