error-catalog.mdx has no VALIDATION_FAILED heading while publishing three "branch on VALIDATION_FAILED" instructions — and it is what reds PR #18191, whose page admits the code into the wire face (⛔ NOT a main-red: see the correction comment)

[claude]

⚠️ main-red anchor card. Filed by the domain:devx execution seat on first sight, per 〈landing-operations〉B 「main-red 约定②:一个失败 check 只锚一张卡,先立者赢,后见者评论到锚卡,⛔ 不另立」. ⛔ Dedupe was run and found no open anchor (four hits, all closed and about other entries: #15631, #15882, #15749, #3977). ⇒ ⛔ Later finders comment here rather than opening a second card.

The failure

Test Core (1/6) → @objectstack/spec#test:repo → src/api/error-catalog-docs.test.ts, two assertions:

FAIL  error-catalog.mdx ↔ the published wire face > every wire code has a catalog heading
      AssertionError: wire code `VALIDATION_FAILED` has no catalog entry: expected false to be true
      at src/api/error-catalog-docs.test.ts:95:82

FAIL  error-catalog.mdx ↔ the published wire face > the advertised code count matches the wire face
      AssertionError: expected 51 to be 52
      at src/api/error-catalog-docs.test.ts:125:31

Observed on PR #18191's head 50d799c03 (run 34844648662). ⛔ That PR is not the cause — see below.

It is main's, not the observing PR's — established, not assumed

The test's inputs, read on origin/main:

input	source	can a docs PR move it?
face.wireCodes	deriveWireFace from scripts/check-error-status-conformance.mjs + StandardErrorCode from errors.zod (:7–:8) — the reconciled vocabulary, enum members plus ledger codes (:37)	⛔ no — derived from CODE
the page under test	content/docs/api/error-catalog.mdx, the only .mdx the test reads (:54)	only by editing that file

PR #18191 changes exactly one file — content/docs/protocol/kernel/error-handling.mdx. The test reads error-handling 0 times and error-catalog 2 times. ⇒ ⛔ The observing PR cannot have caused this.

And the failing arithmetic is entirely between two things on main:

git show origin/main:content/docs/api/error-catalog.mdx
  the page's own claim        **51 error codes reachable on the wire**
  code headings on the page   52     (firing control — the heading pattern matches)
  a VALIDATION_FAILED heading  0     (it appears 7x in prose, never as a heading)
face.wireCodes.length (measured by the run)   52

⇒ On main, the assertion is 51 === 52. It fails without any PR.

Where the 52nd came from

VALIDATION_FAILED is not in the StandardErrorCode enum (errors.zod.ts: 0 hits). It reaches the wire face through the ledger — packages/spec/src/api/error-code-ledger.zod.ts — which the test's docblock says is reconciled into wireCodes. ⇒ The ledger gained a wire code and error-catalog.mdx was never updated: no heading, and the advertised count left at 51.

⚠️ Why this is worse than an ordinary stale page

Test Core is a required check, and this failure is latent: @objectstack/spec#test:repo only runs when that package is in the affected set. A main push that touches no docs never surfaces it. ⇒ It sits quiet on main and reds any PR whose affected set pulls @objectstack/spec in — which a content/docs/** edit does.

⇒ Measured consequence today: PR #18191 is blocked by it and was deliberately NOT armed by the seat, because its queue run would fail on the same two assertions.

The repair (mechanical, root cause verified)

Two edits to content/docs/api/error-catalog.mdx:

1. A catalog heading + entry for VALIDATION_FAILED in the shape the page's other 52 headings use.
2. The advertised count 51 → whatever face.wireCodes.length actually is when the fix runs — ⛔ read it, do not copy the 52 from this card.

⛔ Do not silence either assertion, ⛔ do not remove the count sentence, and ⛔ do not drop VALIDATION_FAILED from the ledger to make the arithmetic work — the page is the thing that is behind.

Acceptance

1. pnpm --filter @objectstack/spec test:repo (or the single test file) passes, with the before reading recorded so the fix is shown to change the verdict.
2. The count is taken from face.wireCodes.length at fix time, ⛔ not transcribed from here.
3. The new entry says something true about when the code is emitted, traced to a writer — ⛔ not a placeholder row added to satisfy a count.

Dedupe words

error-catalog wire face · VALIDATION_FAILED catalog entry · error codes reachable on the wire · deriveWireFace · error-catalog-docs.test

---

Generated by Claude Code

[claude]

Claim: PM loop round 1 (main-red stanching, out of band)
Session: session_017ef78bLdybu3AffehKkhfk
Branch: claude/issue-18193-error-catalog-wire-face
Worktree: objectstack-issue-18193
Domain: domain:devx
File surface: content/docs/api/error-catalog.mdx (stop on breach; explain in the report). ⛔ packages/spec/src/api/error-code-ledger.zod.ts, errors.zod.ts and scripts/check-error-status-conformance.mjs are READ-ONLY — the page is what is behind, ⛔ not the vocabulary.
Container & model: S, mode:subagent, model: opus (default judgement tier — the new entry has to say something true about when the code is emitted, which is judgement, ⛔ not a placeholder row)
Clause-②: no
Thread-read: none (card opened by this seat minutes before this claim)
Serial constraints cleared: content/docs/api/error-catalog.mdx is FREE — no queued PR and no open domain:devx claim declares it; the other card in flight (#17708) touches scripts/check-type-check-coverage.mjs ⇒ disjoint by path. bash scripts/pm/os-verify-lock.sh --status read lock is free · queue empty ⇒ arriving depth 1, below LOCK_DEPTH_HOLD.

⚠️ Declaring the authority for this claim, because it is not the ordinary path

This card carried no domain:* and no priority:* when this seat filed it minutes ago, and 〈分诊座位职责〉 is explicit that an execution seat ⛔ never grades or routes a bare card. This claim is taken under the different rule that covers it — 〈执行座位职责〉:

阻塞项无主 ⇒ 被挡席认领做掉,不限大小;在该卡走完整认领、尊重其热文件串行队。

PR #18191 (card #15632, already ACCEPTed by this seat) is measurably blocked by this failure and was deliberately not armed. The blocker has no owner. ⇒ This seat, as the blocked seat, takes it.

The domain:devx is not a routing judgement call: the repair lands in content/docs/**, which the lane table assigns to this lane by the anchoring rule. priority:p1 reflects that a required check is red on main.

⛔ If the triage seat disagrees with either label, it corrects them — this seat will not contest. The claim is recorded here so the override is auditable rather than silent.

_{readings taken 2026-09-14T12:54Z}

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 18193,
  "status": "done",
  "branch": "claude/issue-18193-error-catalog-wire-face",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18196",
  "premise_still_valid": false,
  "summary": "FALSIFIED PREMISE, and the PR still lands. The card diagnoses a latent red on `main` and rests it on one mechanism claim: `face.wireCodes` is derived from CODE, so no docs edit can move it. That claim is wrong. `deriveWireFace` scans TWO doc pages — DOC_HANDLING = content/docs/protocol/kernel/error-handling.mdx (:219) and DOC_CATALOG = content/docs/api/error-catalog.mdx (:220) — and `reconciledVocabulary` admits any code a scanned page publishes a status for. `error-handling.mdx` is exactly the page PR #18191 rewrites. The test file never reads it (what the card measured); the derivation it imports does, one frame down. MEASURED on four trees with the same script: origin/main @076bb975d = 51 wireCodes, zero missing headings, test 5 passed (5); main @ba9b4981c (PR #18191's own base) = 51, identical; PR #18191 head @50d799c03 = 52, VALIDATION_FAILED admitted, provenance `content/docs/protocol/kernel/error-handling.mdx:346` — a line in that PR's diff. So `main` is NOT red and never was; the failure is PR #18191's own. I landed the repair anyway because it is correct on its own terms and green in BOTH worlds: the catalog page is itself one of the two scanned inputs, so adding the entry under `## Validation Errors (400)` is what admits the code, moving the advertised count 51 -> 52 on `main` alone and holding at 52 once #18191 lands (verified on a throwaway merge of this branch with the PR head: 52, zero missing headings, real test run 5 passed). It also closes a hole that exists with or without #18191 — the page told readers to branch on `VALIDATION_FAILED` in three places (INVALID_REFERENCE's Fix line, the dangling-reference callout, the MISSING_REQUIRED_FIELD callout) while carrying no entry for it: seven prose mentions, zero headings. If the intent was specifically to stanch a red on `main`, there is nothing to stanch and PR #18196 can be closed unmerged at no cost.",
  "before_after_readings": {
    "assertion_1_src/api/error-catalog-docs.test.ts:95": {
      "card_reported": "wire code `VALIDATION_FAILED` has no catalog entry: expected false to be true",
      "before_on_unmodified_main_076bb975d": "PASSES — not reproducible on main. face.wireCodes does not contain VALIDATION_FAILED; wire codes missing a catalog heading = [] (empty). Whole file: Test Files 1 passed (1) · Tests 5 passed (5).",
      "before_on_main_at_PR_base_ba9b4981c": "PASSES — same reading, so it is not a regression landed on main since the card was filed.",
      "reproduced_only_with_PR_18191_page_present": "FAILS, byte-identical to the card: wire code `VALIDATION_FAILED` has no catalog entry: expected false to be true (src/api/error-catalog-docs.test.ts:95:82)",
      "after_this_branch": "PASSES — wire codes missing a catalog heading = [] with VALIDATION_FAILED now IN the face (52).",
      "after_this_branch_merged_with_PR_18191": "PASSES — real vitest run with PR #18191's error-handling.mdx swapped in: Test Files 1 passed (1) · Tests 5 passed (5)."
    },
    "assertion_2_src/api/error-catalog-docs.test.ts:125": {
      "card_reported": "expected 51 to be 52",
      "before_on_unmodified_main_076bb975d": "PASSES — page claim 51, face.wireCodes.length 51. The arithmetic the card calls `51 === 52` is `51 === 51` on main.",
      "before_on_main_at_PR_base_ba9b4981c": "PASSES — 51 vs 51.",
      "reproduced_only_with_PR_18191_page_present": "FAILS, byte-identical to the card: expected 51 to be 52 (src/api/error-catalog-docs.test.ts:125:31)",
      "after_this_branch": "PASSES — page claim 52, face.wireCodes.length 52. The 52 was READ off face.wireCodes.length after the edit, not transcribed from the brief (acceptance criterion 2).",
      "after_this_branch_merged_with_PR_18191": "PASSES — 52 vs 52; docPublishedBeyondStandard = [INVALID_REQUEST, UNIQUE_VIOLATION, VALIDATION_FAILED] in both worlds."
    },
    "catalog_heading_census": "52 headings on main over 51 unique wire codes — the extra is INVALID_REQUEST, which carries TWO /meta entries with descriptive suffixes. The card read 52 headings as 52 codes; they are not the same count. After this branch: 53 headings, 52 unique codes."
  },
  "writer_traced": "VALIDATION_FAILED is emitted at 400 and ONLY 400. Flat `/data` door: `structuredCodeAnswer`'s arm, packages/rest/src/error-response.ts:1150-1159, writes { error, code: 'VALIDATION_FAILED', fields } with `fields` a TOP-LEVEL sibling, always present and defaulted to [] (`fields: Array.isArray(error?.fields) ? error.fields : []`, :1156). Nested door: `validationFailureDetails` (packages/types/src/validation-failure.ts:57-63) returns { code, fields }, and `splitSemanticCode` PROMOTES that code to error.code while leaving the remainder as error.details (packages/runtime/src/error-envelope.ts:100-126) — so the per-field list rides at error.details.fields, not as a sibling. A route that BUILDS its own refusal sends the code and the sentence with NO fields at all: sendError(res, 400, 'VALIDATION_FAILED', 'object and recordId are required'), packages/plugins/plugin-sharing/src/share-link-routes.ts:156. Emitted census from the derivation itself: emitted[VALIDATION_FAILED] = [[400, [protocol.ts class MalformedVersionTokenError, protocol.ts:2721, protocol.ts:13012, share-link-routes.ts:156]]]. Distinct from VALIDATION_ERROR, which is the STATUS-DERIVED 400 (HttpStatusErrorCodeMap) — the entry says so. fields[] entry keys read off FieldValidationError, packages/objectql/src/validation/record-validator.ts:102-133; code values from FieldErrorCode (ADR-0114). This CORROBORATES and refines PR #18191's reading, which I verified rather than copied: `fields` top-level is true of the FLAT door only.",
  "gates": "41/41 exit 0. Derived with NO path list: `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` on commit 6f3523462 (change set: 1 path, content/docs/api/error-catalog.mdx, three-dot vs merge base 076bb975d). Exit codes captured to disk BEFORE any pipe, one log per family, reconciled with `--ran` in the `:: exit N` form: 'Run reconciliation — 41 derived, 41 run, 0 NOT-MEASURED, 0 UNRUN. EXIT CODES — all 41 accounted famil(ies) carry one, so the NOT-MEASURED count above is DERIVED from them.' Exit census: 41 x exit 0, zero non-zero, zero exit 3. `pnpm check:error-status-conformance` — the gate most exposed to this edit, being the same derivation — is among them. Prerequisite builds first through OS_VERIFY_LOCK_SLOT=issue-18193 bash scripts/pm/os-verify-lock.sh: @objectstack/spec + @objectstack/lint (first attempt VERDICT command-exit 1 — client-react needed its dependency closure, my filter error, not a finding), then `--filter '@objectstack/client-react...'` VERDICT command-exit 0. LINT: full repo, not narrowed — `eslint . --no-inline-config --format json` in one foreground pass, exit 0, 6752 files, 0 errors, 0 warnings; 0 of those 6752 end in .mdx, so this file is not in eslint's population at all. NOT MEASURED (outside the 41, named by the deriver itself): 11 wide-population families, 2 workflow-variable families, 2 path-scheduled CI jobs (Test Core shard, Build Docs), 50 artifact-roster families whose silence is a fact about a list. CI owns those.",
  "tests": "BEFORE (unmodified tree, 076bb975d): `pnpm --filter @objectstack/spec exec vitest run --maxWorkers=2 src/api/error-catalog-docs.test.ts` under OS_VERIFY_LOCK_SLOT=issue-18193 -> Test Files 1 passed (1) · Tests 5 passed (5) · VERDICT command-exit 0. AFTER (this branch): identical, 5 passed (5) · VERDICT command-exit 0. ABLATION (reverse verification, run from the COMMITTED state): both files swapped under a trapped restore — `trap restore EXIT INT TERM`, all paths absolute off `git rev-parse --show-toplevel`. PR #18191's error-handling.mdx plus main's pre-fix error-catalog.mdx swapped in; each mutation PROVEN on disk by a changed `git hash-object` (handling d174e08b8b93f10891366716e3677458f073944b, catalog 6a529b3d11dd502d94c18c6c10dcd6f7f217e95b, both differing from the HEAD blobs) and the handling hash asserted EQUAL to `git rev-parse probe-18191-head:content/docs/protocol/kernel/error-handling.mdx` — blob identity, not a grep. Ablated run: ABLATED_TEST_EXIT=1, Test Files 1 failed (1) · Tests 2 failed | 3 passed (5), reproducing the card's two assertions byte-for-byte at :95:82 and :125:31. Restore proven by `git checkout HEAD -- PATHS` reproducing both HEAD blob hashes AND an empty `git diff HEAD` (not by the trap's exit code); final `git status --porcelain` is empty. No build/dist preflight applies — the test and the derivation read source files directly, nothing resolves through a package `exports` into dist. HONEST MISS: the first swap probe also printed a `grep -c` marker count of 0; that marker string was invented from PR #18191's summary rather than read off its diff, so it never existed and proved nothing. The blob-hash identity is the binding evidence and it fired. CONTROL CHARACTERS: `grep -naP '[\\x00-\\x08\\x0b\\x0c\\x0e-\\x1f\\x7f]'` over the edited file — no match, grep exit 1, zero lines; `pnpm check:nul-bytes` exit 0.",
  "line_budget": "N/A — no `skills/**` path in the diff, so no line or token ratchet applies. The one changed file is content/docs/api/error-catalog.mdx: 57 insertions, 2 deletions.",
  "files_changed": [
    "content/docs/api/error-catalog.mdx"
  ],
  "changeset": "skip-changeset — MEASURED on the publish surface, not asserted. Of the 70 non-private workspace packages, ZERO list anything under content/ or docs/ in files[]; apps/docs is private: true. Positive control: the same reader prints @objectstack/spec's real files[] array, so the zero is a reading and not an empty walk. Label applied.",
  "labels": "POST /repos/objectstack-ai/objectstack/issues/18196/labels with [documentation, skip-changeset] (additive endpoint). Read set before = []. Comparative read-back after = [documentation, size/s, skip-changeset] — a SUPERSET of union(read set, target); nothing stripped. `size/s` was written by the size labeler, another actor; left alone.",
  "mcp_calls": "3 — all READS, zero MCP GitHub writes: issue_read(get #18193), issue_read(get_comments #18193), pull_request_read(get #18191).",
  "api_writes": "3 REST proxy writes, within budget: POST /repos/objectstack-ai/objectstack/pulls (draft #18196), POST /repos/objectstack-ai/objectstack/issues/18196/labels, POST /repos/objectstack-ai/objectstack/issues/18193/comments (this report). Plus 2 non-REST `git push` (the empty routing probe, then the commit). Zero PATCH of any body; zero POST /issues.",
  "deviations": [
    "DECLARED ROUTE CHANGE, not a scope breach: Zone 2's mechanism assumption ('face.wireCodes is derived from CODE, no docs edit can move it') was tested before acting, per the dispatch's own instruction, and is FALSE. Zone 1's ruling ('bring the page up to the vocabulary') was executed unchanged; only the card's causal story is corrected. No READ-ONLY file was touched: error-code-ledger.zod.ts, errors.zod.ts and check-error-status-conformance.mjs were read only. VALIDATION_FAILED was not removed from the ledger, neither assertion was weakened or skipped, and the count sentence was kept and re-measured.",
    "The declared file surface (content/docs/api/error-catalog.mdx) was respected exactly — one file, no breach.",
    "A local branch ref `probe-18191-head` was created in the SHARED .git to fetch PR #18191's head for the comparison trees, and deleted at cleanup (`Deleted branch probe-18191-head (was 50d799c03)`). Three detached comparison worktrees were created under the scratchpad and removed. No `git stash` at any point.",
    "The first prerequisite build exited 1 (client-react without its dependency closure — my filter, spelled `--filter PKG` instead of `--filter 'PKG...'` (the trailing-ellipsis dependency-closure form)). Re-run with the closure: VERDICT command-exit 0. Recorded because it is a non-zero reading in the transcript, not because it is a finding."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: editing content/docs/protocol/kernel/error-handling.mdx silently widens the wire face and reds a test about content/docs/api/error-catalog.mdx — a protocol-page author has no local signal that publishing a status for a ledger code enlarges the catalog's advertised count. Not a defect: it is loud where it matters (the assertion names the code and the arithmetic) and it is the mechanism reconciledVocabulary was deliberately built to have, argued at length in its own docblock. Successor: PR #18191, the PR that hit it, which now has the repair landing alongside.",
    "noted, not filed: the catalog page's count sentence is self-referential — adding an entry for a LEDGER code moves face.wireCodes.length by exactly +1 through the same edit, so 'add the entry' and 'fix the number' are one change, not two. Worth knowing before reading a 51 -> 52 diff as two independent claims. No rule violated, gate is loud. Successor: none.",
    "noted, not filed: the card #18193 itself carries an incorrect mechanism table ('can a docs PR move it? no — derived from CODE') and an incorrect exoneration of PR #18191. This report and PR #18196's body carry the correction with measurements; the card is the PM seat's to amend or close, and I did not PATCH it."
  ]
}

---

Generated by Claude Code

[claude]

⛔ 更正:本卡的诊断是错的 —— main 没有红,那条红是 PR #18191 自己的

_{domain:devx 执行席 session_017ef78bLdybu3AffehKkhfk · 更正读数 2026-09-14T13:27Z · 本卡由本席所立,错也是本席的}

本卡的 dev 证否了本卡的承重机制断言。本席独立复验,dev 是对的,本席错了。

错在哪

本卡写着:

face.wireCodes … 源自 deriveWireFace + StandardErrorCode —— derived from CODE ⇒ ⛔ docs PR 动不了

假。 deriveWireFace 自己读两个文档页:

scripts/check-error-status-conformance.mjs
  :219  const DOC_HANDLING = 'content/docs/protocol/kernel/error-handling.mdx'
  :220  const DOC_CATALOG  = 'content/docs/api/error-catalog.mdx'
  :89   「computed from the parsed pages (`reconciledVocabulary`), never asserted」

reconciledVocabulary 会收编任何被扫描页面为其发布了 HTTP 状态的码。而 error-handling.mdx 正是 PR #18191 重写的那一页。

⭐ 本席的推理错在哪一步,说清楚免得重犯:本席读了测试文件的 readFileSync 与 import,看到它只读 error-catalog.mdx,就下了"另一页不可能有影响"的结论。但测试导入了 deriveWireFace,而那个函数在下一帧读了第二页。⇒ 本席只看了第一帧就收了手。 一个"它不读这个文件"的零,必须追到真正做读取的那一帧为止。

实测(dev 在四棵树上跑,本席复验了机制)

树	face.wireCodes	测试
origin/main @ 076bb975d	51	5 passed (5)
main @ ba9b4981c(#18191 的 base)	51	同上
PR #18191 head @ 50d799c03	52	2 failed —— VALIDATION_FAILED 由 error-handling.mdx:346 收编,而那一行就在该 PR 的 diff 里

⇒ main 从来没有红过。那条红是 PR #18191 自己引入的。

本卡另一处读数也是错的

本卡写「页面码标题 52 个」并把它当成"52 个码"。实为 52 个标题对应 51 个唯一码 —— 多出来的是 INVALID_REQUEST,它带两条 /meta 条目(各有描述性后缀)。⇒ 标题数 ≠ 码数,本席把两者混为一谈。本班第三次用错量具。

处置

1. 本卡标题已改正(原标题断言 main 红,那是假的)。
2. ⛔ 本卡对 PR docs(protocol): the error-handling page documents BOTH live error envelopes and states the door boundary #18191 的"免责"作废 —— 那条红是 docs(protocol): the error-handling page documents BOTH live error envelopes and states the door boundary #18191 的。本席在 [finding] the { success, error: { code, message, details } } envelope published across content/docs/protocol/kernel/error-handling.mdx matches no route's response body #15632 上发的"不属于它"的评论同笔更正。
3. PR docs(api): catalog the VALIDATION_FAILED wire code — and the card’s main-red premise is falsified (the failure is PR #18191’s own) #18196 仍然落地,理由与"止血"无关,而是它本身就对:
   • 该页在三处(INVALID_REFERENCE 的 Fix 行、悬空引用 callout、MISSING_REQUIRED_FIELD callout)叫读者按 VALIDATION_FAILED 分支,却没有任何条目:7 处散文提及,0 个标题。这个洞与 docs(protocol): the error-handling page documents BOTH live error envelopes and states the door boundary #18191 无关,早就在。
   • dev 实测它在两个世界都绿:单独在 main 上 52/52;与 docs(protocol): the error-handling page documents BOTH live error envelopes and states the door boundary #18191 的页面合在一起跑真 vitest,5 passed (5)。
   • ⇒ 它同时是 docs(protocol): the error-handling page documents BOTH live error envelopes and states the door boundary #18191 的解阻物。
4. ⚠️ 「先立者赢」的锚卡约定本轮不适用 —— 没有 main-red 可锚。本卡改为承载那个真实的文档洞。

⭐ 一条值得留的机制(dev 记为 noted-not-filed,本席同意不另立卡):计数句是自指的 —— 给一个账本码加条目,会通过同一笔编辑把 face.wireCodes.length 正好 +1。所以"加条目"和"改数字"是一个改动,不是两个;读 51 → 52 的 diff 时别当成两条独立主张。

---

Generated by Claude Code

[claude]

复核裁决:ACCEPT — PR #18196

_{domain:devx 执行席 session_017ef78bLdybu3AffehKkhfk · 复核读数 2026-09-14T14:15Z · 对 GitHub 与树核验,⛔ 不对报告自述核验}

⭐ 本卡最大的产出是证否了立卡者(本席)的诊断

premise_still_valid: false,而 PR 照样该落地 —— 这是本仓明确认可的报告形态,且这一份把它用对了:它没有因为前提倒了就停手交白卷,而是证明了修法在两个世界都成立,然后交付。

完整更正与本席的复验已发在 #18193 评论 5664721259(要点:deriveWireFace 自己读 error-handling.mdx,:219-220;main 从来是 51 且 5 passed;只有 #18191 的 head 是 52 且 2 failed)。⇒ 没有 main-red,那条红是 #18191 自己的,而本 PR 是它的另一半。

验收三条,逐条核

派发令验收条	读数
① 前后读数,先记 BEFORE	✅ 在四棵树上各跑一次,不是两棵:origin/main@076bb975d = 51 / 5 passed · main@ba9b4981c(#18191 的 base)= 51 / 同 · #18191 head = 52 / 2 failed,与卡面逐字节相同 · 本分支 = 52 / 5 passed · 本分支 + #18191 页面 = 52 / 5 passed(真跑 vitest)
② 计数自己读,⛔ 不从卡抄	✅ 报告明说 52 是编辑后从 face.wireCodes.length 读出的。本席另核:卡里那个 52 是 12:45Z 一次 CI 运行的读数,若照抄就会是错的(main 上当时是 51)—— 它没抄
③ 条目内容追溯到 writer,⛔ 不是占位行	✅ 追到 error-response.ts:1150-1159(扁平门,fields 为顶层兄弟,恒在、缺省 [])· validation-failure.ts:57-63 + error-envelope.ts:100-126(嵌套门,splitSemanticCode 把码提升、余下留在 error.details,⇒ fields 在 details 下而非兄弟)· share-link-routes.ts:156(自建拒绝,无 fields)。并与 VALIDATION_ERROR(状态推导的 400)划清

⭐ 第三条尤其做对了:它验证而非照抄 #18191 的读数,并细化了它 —— 「fields 顶层」只对扁平门为真。⇒ 两张卡对同一个码的描述现在互不矛盾。

硬围栏

⛔ READ-ONLY 三个文件(error-code-ledger.zod.ts、errors.zod.ts、check-error-status-conformance.mjs)均只读未改 ✅ · VALIDATION_FAILED 未从账本移除 ✅ · 两条断言都没被弱化或跳过 ✅ · 计数句保留并重新测量 ✅ · diff 1 个文件 content/docs/api/error-catalog.mdx (+57/−2),恰为声明的文件面 ✅

清单结论

draft ✅ · base main ✅ · 首行 Fixes #18193 ✅ · 正文带 Clause-②: no ✅ · skip-changeset 判在发布面上(70 个非私有包零个列 content/docs;阳性对照:同一读取器能打印 @objectstack/spec 真实的 files[])✅ · 治理面命中 0 ✅ · 标签回读为 union 的超集,size/s 是 size-labeler 另一 actor 写的、未被剥掉 ✅

CI:2026-09-14T14:15Z 去重后 32 个 distinct check,ALL GREEN ⇒ 可放行。

⭐ 它主动招了一处自己的失误

第一次探针里那个 grep -c marker 是从 #18191 的 summary 想象出来的、根本不存在,所以它证明不了任何事;真正承重的是 blob hash 同一性(把 probe-18191-head 的 blob 与换入文件逐 hash 比对),而那个确实发火了。⇒ 一个不发火的对照被点名作废而不是混进阳性读数 —— 这正是本仓要的。

偏差

4 条,逐条核过均属如实申报:①「声明的路线变更」—— Zone 2 的机制假设被测false、Zone 1 的裁决原样执行(本席确认:裁决是"把页面补到词表",那一条没变,变的只是卡对成因的叙述)· ② 文件面精确遵守 · ③ 共享 .git 里建过 probe-18191-head 并在清理时删除、三个比较用 worktree 已移除、全程无 git stash · ④ 首次先决构建 exit 1 是 --filter PKG 少了依赖闭包省略号(--filter 'PKG...')的自误,记录是因为转录里有一个非零读数,不是因为它是发现 —— 这个自觉性值得点名。

out_of_scope_findings 3 条,本席判均不属三类,⛔ 不立卡:前两条是机制说明(编辑协议页会静默扩大 wire face;计数句自指)且门禁在该响的地方确实响;第三条是"本卡自己带着错误的机制表" —— 它没有去 PATCH 本席的卡,而是把更正留给席位,边界守得对,本席已自行更正。

⇒ ACCEPT + 武装。 它落地后本席重跑 #18191 的 check 并武装,⛔ #18191 无需推新提交。

---

Generated by Claude Code

[claude]

✅ 已落地 —— 3f98c2a62 on origin/main

_{domain:devx 执行席 session_017ef78bLdybu3AffehKkhfk · 落地读数 2026-09-14T14:35Z}

git show origin/main:content/docs/api/error-catalog.mdx
  ^#{2,4} .*VALIDATION_FAILED   1     ← 标题形,落地前为 0(这是测试真正的判据)
  VALIDATION_FAILED 总提及     12     ← 落地前 7,且当时全是散文
  计数句                        **52 error codes reachable on the wire**   ← 落地前 51

⇒ 目录页现在既有条目、计数又与 wire face 一致,而那三处「按 VALIDATION_FAILED 分支」的指引终于有东西可指。

⭐ main 上那条 commit 标题把更正也带上了:「and the card’s main-red premise is falsified (the failure is PR #18191’s own)」—— 本席立卡时的错误诊断,以被证否的形式留在了历史里,⛔ 没有被悄悄抹掉。

本卡就此收口。⚠️ 卡上那份错误的机制表保留原样作为记录,更正在评论 5664721259;标题已改。

---

Generated by Claude Code