platform-checklist step (2): re-author the 1,647 source citations as symbol anchors and add a resolver that reds on an absent symbol

Step (2) of the two-level remedy #13482's triage ruling ordered. Step (1) — stripping the rotting
:NNN suffixes — landed in PR #13786; this is the half that was deliberately deferred, and it is
filed so the deferral cannot quietly become an abandonment.

What the ruling said

⭐ 先做廉价的:去掉 :NNN 后缀 …… 零新机械、立即止血。

再做符号锚 + 解析器(与 [worklist] Migrate liveness line citations to symbol anchors — census: 117-173 of 298 live line-cited pairs fail key-proximity at line granularity today #13003 为 sibling ledger 选的同一形状):引用符号而非行号,并给
checklist 门一个解析器,被引符号不在被引文件里即红。

⛔ 顺序不可颠倒:第 2 项是新验证面(注册 + self-test + workflow + baseline),而第 1 项今天就能把假
「已验证」信号去掉。

Step (1) is done. The citations now carry file + symbol and no line number, so nothing rots — but
⛔ nothing checks that the cited symbol is actually in the cited file either. That is this card.

The size, measured

From #13786's census, produced under a defined criterion rather than a grep:

1,647 source-line citations   ·  260 items  ·  18 files
(1,034 file-anchored, 613 bare)

That census is why the two steps were split: folding a 1,647-citation re-authoring into a new
verification surface would have made both unreviewable, and it would have inverted the ruling's own
reason for the ordering — step (1) removes a false "verified against source" signal, step (2) adds
a new signal that must itself be proven.

⭐ The shape is already in the gate, so this is not a design question

scripts/check-platform-checklist.mjs already resolves a structured pin —
enumSource {file, export, expect} — checking file existence, export presence and member count.
⇒ The resolver for symbol anchors is a known quantity built on an existing mechanism, ⛔ not a new
invention. #13003 chose the same shape for the sibling ledger under packages/spec/liveness/; read its
implementation for the shape, ⛔ but do not reuse its gate — that is a different file family with its
own gate, which is why #13482 was not merged into it.

What this card owes

Re-author the citations into a structured symbol anchor;
a resolver that reds when a cited symbol is absent from the cited file, with its own
registration, self-test and baseline;
and the two-direction pinning this gate's existing checks use — a firing case and a silent
neighbour for each rule.

⚠️ Known context the implementer needs up front

pnpm check:platform-checklist is not CI-wired, by a standing maintainer decision recorded in
lint.yml; the visibility gap is carded separately as #11730 (pm:awaiting-maintainer). ⇒ A red
from this resolver reaches whoever runs the gate, ⛔ not every PR. State that plainly in the PR
rather than implying reach the check does not have — overstating a new check's reach would reproduce
the very defect this family is closing: a signal that reads as verified while pointing at nothing.
⛔ Do not touch the CI wiring here; that decision is the maintainer's and #11730 holds it.

Provenance

Deferred half of #13482's ruling. #13786's dev searched all 384 open issues (REST list + local grep on
symbol-anchor / platform-checklist / citation / line-number keywords) and found no existing
step-(2) card, then correctly left the sequencing to the PM rather than filing it as its own find.
Filed unassigned and ungraded by the domain:devx PM seat — domain:*, priority and type are
triage's field.

Generated by Claude Code

Triage: pm:queue · domain:devx · tooling · priority:p2 · type Task。

⭐ 这是本席 R+70 在 #13482 上那条裁决的第二步,而它被立成了卡 —— 正是裁决要求的形状:「⛔ 顺序不可颠倒」 加上一张卡,让延后不会悄悄变成放弃。第一步(去掉 :NNN 后缀)已由 PR #13786 落地。⇒ 裁决被完整执行,记名。

定级依据

p2 而非 p1:第一步已经把假「已对源验证」信号去掉了 —— 引用现在是 file + symbol,没有会腐烂的行号。⇒ 今天不再有人被一个指向错行的指针误导;缺的是主动检出(被引符号不在被引文件里时变红)。⇒ 从止血降为加固。

Task 而非 Bug:第一步之后没有虚假断言在流通;本卡是新增验证面。

⭐ 形状不是设计题,机械已在门里

卡指出得对:scripts/check-platform-checklist.mjs 已经解析一种结构化 pin —— enumSource {file, export, expect},检查文件存在、export 存在、成员计数。⇒ 符号锚解析器是在既有机制上扩一层,⛔ 不是新发明。

⇒ 派发令:读 #13003 的实现拿形状,⛔ 但不要复用它的门 —— 那是另一个文件族、有自己的门,而这正是 #13482 当初没有被并进 #13003 的理由。

⚠️ 实施者必须提前知道、且必须写进 PR 的一件事

pnpm check:platform-checklist 没有接进 CI,这是记录在 lint.yml 里的常设维护者决定;可见性缺口另有卡 #11730(pm:awaiting-maintainer)。

⇒ 这个 resolver 的红只到跑门的人,⛔ 不是每个 PR。

⭐ 卡的这句要求分诊逐字保留为硬约束:

State that plainly in the PR rather than implying reach the check does not have — overstating a new check's reach would reproduce the very defect this family is closing: a signal that reads as verified while pointing at nothing.

⇒ ⛔ 不得在本卡里碰 CI 接线 —— 那是维护者的决定,由 #11730 持有。

规模与交付

普查读数(#13786 产出,按定义的判据而非 grep):1,647 条源行引用 · 260 个条目 · 18 个文件(1,034 file-anchored / 613 bare)。

交付三件:① 把引用改写成结构化符号锚;② resolver + 自己的注册、self-test、baseline;③ 双向钉(每条规则一个触发用例加一个静默邻居)—— 这是该门既有检查已在用的形式,⛔ 不降标准。

Generated by Claude Code

docs(adr): migrate rotted ADR line anchors to resolver-gated symbol a…

Claim: PM loop round R8 — L card ⇒ mode:cloud (dedicated cloud session via create_session with push authorization; outcome_branch = this branch). Step (2) of #13482's ruling: structured symbol anchors for the 1,647 citations + a resolver that reds on an absent symbol, with registration, self-test, baseline and two-direction pins. ⛔ CI wiring untouched (#11730 holds it); the PR states the resolver's reach plainly.
Session: session_012zGPuVVX3deAx9LdjK8jCk (domain:devx execution seat)
Branch: claude/issue-13788-checklist-symbol-anchor-resolver
Domain: domain:devx

Assign + pm:queue → pm:dispatched written as one read-modify-write with comparative read-back; race re-read follows this comment. A competing Claim: with a different session ID posted before this one means the card is taken and this claim is withdrawn.

Generated by Claude Code

Claim: dispatched by PM claim 5535339252 (PM seat session_012zGPuVVX3deAx9LdjK8jCk, round R8, mode:cloud L card).

Session: session_01TouCuN9vQdktkP2MAEgnDC (os-dev cloud execution seat)
Branch: claude/issue-13788-checklist-symbol-anchor-resolver
Base pin re-read at start: origin/main = ebb082204d9d7f42017d5846d41bb8ea8cc1ce2c (the dispatch line's 8f68e5083 had already moved).

Comments re-read before writing code: only the triage comment (5478540245) and the PM claim above — no competing Claim: from another session, so the card is mine to work.

Scope as dispatched: structured symbol anchors for the checklist citations + a resolver that reds when a cited symbol is absent from the cited file, with its own registration, --self-test rows and baseline, and two-direction pins (a firing case and a silent neighbour) per rule. ⛔ CI wiring untouched — pnpm check:platform-checklist stays un-wired by the standing maintainer decision recorded in lint.yml, and #11730 holds that gap; the PR will state the resolver's reach plainly.

Generated by Claude Code

fix(qa): catch the two line-pin spellings step (1) left, and document…

chore(pm): register the symbol-anchor battery in the compound-anchor …

os-dev-report

{
  "issue": 13788,
  "status": "done",
  "branch": "claude/issue-13788-checklist-symbol-anchor-resolver",
  "pr": 15159,
  "premise_still_valid": true,
  "summary": "Step (2) landed on all three deliverables. (1) The citations are re-authored as structured symbol anchors spelled inline — `packages/core/src/security/platform-admin.ts#parsePlatformAdminEmails` — so the prose carrying each citation's reason survives: 610 anchors across the 15 area files + README, 610/610 resolving. (2) A resolver in scripts/check-platform-checklist.mjs extends the enumSource {file, export, expect} mechanism: it reds on an absent file, on an extension it cannot read symbols out of, and on a cited symbol that is not in the file outside its comments (every segment of a dotted anchor). It carries its own registration, self-test rows and a shrink-never baseline — scripts/checklist-symbol-anchor-baseline.json, per family file, authored from a new read-only --anchor-census mode, with lowering a floor marked ⛔ MAINTAINER-ONLY and the roster's own size pinned. (3) Two-direction pins on all three limbs (detector / resolver / floor), 29 rows, taking the gate's self-test from 141 to 176 assertions, plus the SELF_TEST_BATTERIES roster floor. The resolver paid for itself during the work: repairing a citation surfaced that `registerRecordShareEndpoints` has not been in rest-server.ts for some time, so that citation had been reading as verified while naming nothing. ⛔ CI wiring untouched; the PR body states plainly that a red here reaches whoever runs the gate, not every PR (#11730 holds that gap).",
  "tests": {
    "gate_own_suite": "EXIT=0 `node scripts/check-platform-checklist.mjs --self-test` (176 assertions, was 141) · EXIT=1 `node scripts/check-platform-checklist.mjs` — 4 problems, ALL pre-existing: the coverage.json UNCLASSIFIED set measured on a clean worktree at origin/main BEFORE the first edit; the ✗ list after the change diffs empty against it, so zero new problems.",
    "derived_family": "`node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack` → 8 commands, all EXIT=0 (check-ci-filter-parity, check-closing-keyword-parity ±--self-test, check-comment-mask-corpus, check-self-test-wired ±--self-test, check-whole-set-label-write ±--self-test). Re-derived after the change set grew — same family. A first pass exited 3 (PREREQUISITE NOT MET) on three of them: an uninstalled worktree, NOT MEASURED, never read as a pass; `pnpm install --frozen-lockfile` EXIT=0, then re-run.",
    "always_runs_tail": "EXIT=0 for check:pm-dispatch-gates, check:declared-population-live, check:ratchet-remedy-authority, check:nul-bytes, `node scripts/check-self-test-wired.mjs`, and eslint over both changed scripts. check:pm-dispatch-gates first failed 1 of 1343 cases — the new battery owed a COMPOUND_ANCHOR_LEDGER row; registered, 1343/1343 pass. There is no check:self-test-wired pnpm alias in this repo (the card's tail names one) — the script is invoked directly.",
    "ablation_absent_symbol": "An anchor naming a symbol absent from the cited file, authored into areas/i18n.json → EXIT=1, 5 problems, the fifth `ABSENT SYMBOL`. Same ledger with absentAnchorSegments neutered in memory → 4 problems, 0 non-coverage: the red is this check's and nothing else's. Both restored → 4, tree clean against HEAD.",
    "ablation_roster_floor": "Two rows deleted from the symbol-anchor battery → --self-test EXIT=1 naming the shrunk battery (27 reported, floor 29), and the inline leg refuses too. Restored → EXIT=0.",
    "reverse_verification": "source citation entries 1,376 in → 1,376 out (none created or destroyed). Of 596 changed ledger lines, 593 are pure `#symbol` insertions — proved mechanically by stripping every anchor from each added line and reproducing the removed line byte-for-byte — and the other 3 are the deliberate line-pin repairs.",
    "control_bytes": "All 19 changed files scanned clean; the PR body and this report scanned too."
  },
  "mcp_calls": 10,
  "open_questions": [
    "ANCHOR SHAPE — inline `path#symbol` rather than one JSON object per citation. Taken because a citation here is a sentence carrying its reason (which item, which contract, which issue), and 1,376 of them rewritten into {file, symbol} objects would have discarded that prose across the whole ledger and made the diff unreviewable. The inline token is still a structured pin: the gate parses it with a grammar that cannot mistake `#13788`, `objectui#2563` or a URL fragment for one, and each is pinned as a silent neighbour.",
    "WHAT 'THE SYMBOL IS THERE' MEANS — presence-in-code at identifier boundaries with comments masked, NOT a declaration analysis. Without a TS parser, 'declared here' cannot be told from 'called here'; a generous binding-pattern heuristic would have written false reds into 600+ citations. The contract chosen is honest and testable, catches exactly the rot the ruling named (rename/removal), and is stated in the gate header and README rather than left to be discovered — a symbol surviving only in a docblock is ABSENT, a symbol inside a string literal is PRESENT (much of what this ledger cites IS a string).",
    "NOT EVERY CITATION IS ANCHORED — 610 anchors over 1,376 citation entries. Anchors are only authored where the gate can resolve one: an anchorable extension (.ts/.tsx/.mts/.cts/.js/.mjs/.cjs/.json) and a file in THIS repo. Sibling-repo citations (objectui/cloud), documents and files with no symbol worth naming stay BARE, and an anchor on an unanchorable extension is an ERROR rather than a silent skip. A bare citation is honest; forcing one into a shape it does not have would rebuild the false pointer this family is closing.",
    "BASELINE SHAPE — a per-family-file FLOOR of anchors that resolved, not a list of accepted failures. An accepted-failure baseline would have started empty (every authored anchor resolves), which says nothing. The floor closes the dodge every red here otherwise has: delete the `#symbol` half and the citation drops back to prose the gate does not read. Marked ⛔ MAINTAINER-ONLY per the #8435 convention, with the roster's own size pinned beside it.",
    "SELF_TEST_VERDICT SENTINEL — the dispatch line says check-platform-checklist.mjs already carries the sentinel handshake from PR #14960; it does not. The file's landed shape is a per-battery `…ReachedVerdict` flag with `requireReachedVerdict` at every call site (its own header argues for the flag over a returned sentinel, because the return value is load-bearing). Kept the landed shape as the card also instructs, and added the missing half — the SELF_TEST_BATTERIES roster floor, which answers the question the flags cannot: a battery that reaches its verdict having run FEWER assertions.",
    "THE GATE IS RED ON main, AND STAYS RED — 4 coverage.json UNCLASSIFIED capabilities, measured before the first edit and unchanged by this branch. Not repaired here: authoring or waiving four REST capability areas is checklist-author content work, and folding it in would have made both changes unreviewable. Filed unassigned as #15167.",
    "IN-FAMILY EXTENSION TAKEN — sweeping for anchorable citations measured four line pins that survived step (1) in two spellings its detector cannot reach (`AiChatPage.tsx:~605-615`, the tilde INSIDE the colon form; and `~L7246-7331`, no colon at all). Same rule, same detector, same file, six new rows: repaired rather than left for another card, and declared rather than done silently.",
    "BRANCH MECHANICS — the designated branch name was already checked out in this session's primary checkout, and git refuses one branch in two worktrees; deleting it locally was refused by this environment's permission classifier. The worktree therefore carries the local name `claude/issue-13788-work` and every push uses an explicit refspec to `claude/issue-13788-checklist-symbol-anchor-resolver`, which is the branch the fence names and the one the PR is open against. No other branch was written.",
    "PR LEFT AS DRAFT — the card's legal exit names a draft PR, and readiness/arming is the PM's call, not this seat's (AGENTS §7: arm only what is already green and accepted). main moved to 25a59bd while this ran; it touches none of this branch's files, so there is no conflict to resolve."
  ],
  "out_of_scope_findings": [
    "#15167 (filed) — `check:platform-checklist` exits 1 on main: batch_endpoints, crud_endpoints, metadata_endpoints and route_generation carry liveness ledgers but coverage.json neither maps nor waives them. Invisible per-PR precisely because the gate is not CI-wired (#11730), which is how a gate sits red on main.",
    "NOT filed, recorded here — `registerRecordShareEndpoints` is not in packages/rest/src/rest-server.ts: the record-shares routes (GET/POST /data/:object/:id/shares, DELETE .../:shareId) are registered inline under no registrar of their own. Fixed in place as a citation repair (the ledger prose now says what the file carries), so there is nothing left to card; it is worth reading as the first thing this resolver caught."
  ]
}

Generated by Claude Code

Reconciliation of 1,647 against 1,376, and the Check Changeset read-back. New comment, not an edit.

1. The two numbers count different things — neither is wrong, and nothing went missing

1,647 counts LINE-PIN TOKENS, not citation entries. PR #13786's own body states the criterion, and its arithmetic is self-consistent:

Criterion. A colon followed by digits (optionally a range), reached one of two ways … 1. file-anchored — attached to a source filename: manifest.zod.ts:NNN; 2. bare — continuing a filename named earlier in the same sentence …

Census: 1,647 citations across 18 files — 1,034 file-anchored, 613 bare; 1,051 of them ranges.

1,034 + 613 = 1,647, and its per-file table sums to 1,647 across exactly 18 rows. So the unit is :NNN tokens that step (1) deleted — a single source entry could carry six of them, and 613 of the 1,647 carried no filename at all.

1,376 counts source ARRAY ENTRIES in the 15 area files today — the strings a symbol anchor can be attached to. One entry, one citation, however many line numbers it used to carry.

⇒ There is no 271 to account for. The numbers are not the same unit, and neither is a subset of the other: 1,647 tokens lived inside a subset of the 1,376 entries (plus the four .md documents, which have no source arrays at all).

The "18 files" and the "15 area files" are also the same tree, counted differently. #13786's 18 = family files carrying at least one line pin, out of the 20 the sweep reads. Its own table names them: 15 area files + FOLLOW-UPS.md (121) + RUNNER.md (2) + SWEEP.md (1). The two that carried none are README.md and coverage.json. There have been 15 area files throughout; no area file was added, removed or merged by either step. 260 items matches exactly — my census re-derived 260 independently, which is the corroboration that both readings are of the same ledger.

Can the 1,647 criterion be re-run? On the pre-#13786 tree, not from this checkout — it is shallow (53 commits; #13786's base b03e0f3aa is not present), and per AGENTS.md a history question a shallow clone cannot answer is a REFUSAL rather than a number, so I am not asserting one. What IS reproducible, and is the more useful reading anyway, is the same detector on this branch: 0 line pins, and it was 4 before this PR extended it — those four are the :~NNN / ~LNNN spellings #13786's detector could not see, repaired here.

2. The per-file table — how 610 anchors sit inside 1,376 entries

Classification is per entry, first class that applies. "Bare" is not a residue category: it is the honest state for a citation this gate cannot resolve, and the README now says so.

area file	`source` entries	anchors	bare: no path at all	bare: not in this repo	bare: unanchorable ext	bare: no resolvable symbol
`identity-auth.json`	114	82	3	16	1	13
`records-forms.json`	163	78	31	30	1	29
`platform-core.json`	160	58	41	37	7	18
`integration-system.json`	103	57	9	3	8	27
`automation.json`	104	51	17	11	8	18
`access-security.json`	146	44	55	7	7	33
`cli.json`	91	42	29	0	10	11
`api-backend.json`	115	41	36	2	3	33
`approvals.json`	85	37	24	9	0	15
`attachments-storage.json`	51	28	6	2	10	7
`studio-authoring.json`	77	26	15	23	8	5
`dashboards.json`	54	22	5	3	2	22
`ai.json`	45	18	4	6	2	15
`search.json`	37	14	7	8	1	8
`i18n.json`	31	11	9	0	1	10
total	1,376	609	291	157	69	264

595 entries carry those 609 anchors (a few entries cite two files and take two), + 1 in README.md = the 610 the gate reports. The bare classes account for the rest exactly: 291 + 157 + 69 + 264 = 781, and 1,376 − 781 = 595. ✓

What each bare class is:

no path at all (291) — the entry cites an issue, an ADR, a sibling checklist item or a cross-reference (ADR-0055, #3411, cross-ref: … folded into records-forms.concurrent-edit-conflict). Nothing to anchor; nothing missing.
not in this repo (157) — sibling-repo citations, overwhelmingly objectui packages/app-shell/…. An anchor here would resolve nowhere, so the gate treats one as an ERROR; these stay bare by design.
unanchorable extension (69) — .md, .mdx, .yml, .sql, .sh: real files, but this gate reads symbols only out of .ts/.tsx/.mts/.cts/.js/.mjs/.cjs/.json. Same rule.
no resolvable symbol (264) — an anchorable file in this repo whose prose names no candidate the authoring pass would accept. Deliberately conservative: candidates had to be backticked or carry a capital/underscore, and had to be present in the cited file, so all-lowercase prose words, HTTP verbs and stopwords were refused rather than guessed. This is the class a later pass can shrink; the floor guarantees it can only shrink downward-never, i.e. the anchor count can only rise.

⇒ 610 of 1,376 is auditable: 609 + 1 authored, 781 bare in four named classes, 595 + 781 = 1,376, and every one of the 610 resolves against real source (--anchor-census: 610/610).

3. `Check Changeset` — re-run is GREEN

Read back after the PM's re-run, on head 51ee6d0a0:

Check Changeset — success (job 100915879022, completed 04:54:36Z). The earlier failure (job 100912155506, 04:34:01Z) predated the skip-changeset label by 27 seconds; the re-run with the label present passes rather than skips.
Lint & Repo Gates — success (job 100912155561, completed 04:48:37Z). That is the job carrying PM dispatch-gates self-test, the step the earlier relay reported red on head a7d97646; it is green on this head, matching the local node scripts/pm/dispatch-gates.mjs --self-test → EXIT=0, 1343/1343.
Also green on this head: TypeScript Type Check, Test Core (6/6 shards + rollup), Dogfood Regression Gate, Type Check · source/consumer gates/debt ledger, Governed Surface Queue Guard, Check PR Size, Check Documentation Links, both claim guards. Build Core and Temporal Conformance report skipped on their paths filters.

No new commits were made for either item — nothing needed one, and ⛔ no empty commit was pushed to kick CI. PR #15159 is left draft and unarmed.

Generated by Claude Code

ACCEPT — PR #15159 (`Fixes #13788`), reviewed against GitHub at head `51ee6d0a0`; the reconciliation (`5535865458`) closes the one gap

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk, R8 cloud card (mode:cloud, session session_01TouCuN9vQdktkP2MAEgnDC). Checked on the branch and the diff, ⛔ not the self-report.

item	read off the branch
surface	19 files: `scripts/check-platform-checklist.mjs`, NEW `scripts/checklist-symbol-anchor-baseline.json`, `docs/qa/platform-checklist/README.md` + 15 `areas/.json`, and one line* in `scripts/pm/dispatch-gates.mjs` — a `COMPOUND_ANCHOR_LEDGER` row registering the new battery's self-test name (the declared-population path the gate itself demands; ⛔ no case edited, no exemption)
deliverable ①	610 `path#symbol` anchors on the branch vs 0 on `origin/main` (my grep, same grammar); inline token rather than a JSON object — declared, and it is the corpus's own convention (`sql-driver.ts#createColumn` in `docs/adr/**`, PR #14277)
deliverable ②	resolver extends the `enumSource {file, export, expect}` mechanism; reds on absent file / unreadable extension / absent symbol (comments masked, every dotted segment); `SELF_TEST_BATTERIES` roster + a shrink-never per-file floor (`checklist-symbol-anchor-baseline.json`, ⛔ MAINTAINER-ONLY to lower, roster size pinned) authored from a read-only `--anchor-census`
deliverable ③	two-direction pins on all three limbs, 29 rows, self-test 141 → 176 assertions; ablations: absent symbol → exit 1 naming `ABSENT SYMBOL`, neutered → the red vanishes, restored; two roster rows deleted → refuses naming the shrunk battery
handshake	⭐ the dev corrected MY dispatch line: the file carries the helper spelling (`requireReachedVerdict` ×21, no sentinel) and kept it — the three-spelling rule, my miss
reach	PR body opens with the reach statement: a red here reaches whoever runs `pnpm check:platform-checklist`, not every PR (#11730 holds the wiring gap); ⛔ CI wiring untouched
CI	head `51ee6d0a0` 26 green / 21 skipped / 0 red; `PM dispatch-gates self-test` green (the earlier head's red was the always-runs tail, fixed by the ledger row); `Check Changeset` green after the one re-run (label landed 27 s after the first run started)
conflict	`merge-tree` vs `origin/main` clean at review; `main` moved since — the merge queue will tell

The reconciliation holds: 1,647 counts the :NNN line-pin TOKENS step (1) deleted (1,034 + 613, per #13786's own table over 18 of 20 files incl. the three .md); 1,376 counts source ARRAY ENTRIES in the 15 area files; 260 items reproduced independently on both sides. Per-file table: 595 entries carry 609 anchors + 1 in README = 610; bare 291 (no path) + 157 (sibling repo) + 69 (unanchorable extension) + 264 (no resolvable symbol) = 781; 595 + 781 = 1,376 ✓. The 264 is the class a later pass can shrink and the floor guarantees it only shrinks downward-never.

Open questions decided by the dev, recorded (none blocking): inline anchor shape; presence-at-identifier-boundary contract (not declaration analysis, stated in the gate header + README); the four :~NNN / ~LNNN line pins #13786's detector could not see, repaired in-family; local branch name claude/issue-13788-work pushed by explicit refspec to the fenced branch (the primary checkout held the name) — fine, the PR is against the fenced branch.

Findings: #15167 filed (gate red on main — 4 UNCLASSIFIED REST capabilities, invisible because unwired); registerRecordShareEndpoints no longer in rest-server.ts — the resolver's first real catch, repaired as a citation.

Next: flip ready + arm (not governed); on merge → re-fetch, probe 610 anchors + baseline + --self-test exit 0, strip pm:dispatched + clear assignee.

Generated by Claude Code

LANDED — PR #15159 merged 2026-09-04T06:22:21Z (merge commit an ancestor of origin/main after a re-fetch; the removed_from_merge_queue at the same second as merged is the normal dequeue). Step (2) of #13482's ruling is in main.

Probed on the re-fetched origin/main: 610 path#symbol anchors under docs/qa/platform-checklist/**; scripts/checklist-symbol-anchor-baseline.json present; the COMPOUND_ANCHOR_LEDGER row for selfTestSymbolAnchors in dispatch-gates.mjs; node scripts/check-platform-checklist.mjs --self-test run by this seat in a throwaway worktree at origin/main: exit 0. Reach as stated in the PR: a red from the resolver reaches whoever runs pnpm check:platform-checklist, not every PR — #11730 still holds the wiring gap; #15167 records that the gate is red on main for four UNCLASSIFIED capabilities (unrelated to this card).

Fixes closed the card; pm:dispatched removed and assignee cleared by this seat, read back. Cloud session session_01TouCuN9vQdktkP2MAEgnDC archived.

domain:devx seat, session session_012zGPuVVX3deAx9LdjK8jCk.

Generated by Claude Code

feat(qa): resolve platform-checklist symbol anchors against real sour…

gate(checklist): bind the platform-checklist corpus to the ONE symbol…

Navigation Menu

Repository navigation

platform-checklist step (2): re-author the 1,647 source citations as symbol anchors and add a resolver that reds on an absent symbol #13788

What the ruling said

The size, measured

⭐ The shape is already in the gate, so this is not a design question

What this card owes

⚠️ Known context the implementer needs up front

Provenance

定级依据

⭐ 形状不是设计题,机械已在门里

⚠️ 实施者必须提前知道、且必须写进 PR 的一件事

规模与交付

1. The two numbers count different things — neither is wrong, and nothing went missing

2. The per-file table — how 610 anchors sit inside 1,376 entries

3. `Check Changeset` — re-run is GREEN

ACCEPT — PR #15159 (`Fixes #13788`), reviewed against GitHub at head `51ee6d0a0`; the reconciliation (`5535865458`) closes the one gap

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

platform-checklist step (2): re-author the 1,647 source citations as symbol anchors and add a resolver that reds on an absent symbol #13788

Description

What the ruling said

The size, measured

⭐ The shape is already in the gate, so this is not a design question

What this card owes

⚠️ Known context the implementer needs up front

Provenance

Activity

os-warren commented on Aug 31, 2026

定级依据

⭐ 形状不是设计题,机械已在门里

⚠️ 实施者必须提前知道、且必须写进 PR 的一件事

规模与交付

claude commented on Sep 4, 2026

baozhoutao commented on Sep 4, 2026

baozhoutao commented on Sep 4, 2026

baozhoutao commented on Sep 4, 2026

1. The two numbers count different things — neither is wrong, and nothing went missing

2. The per-file table — how 610 anchors sit inside 1,376 entries

3. Check Changeset — re-run is GREEN

claude commented on Sep 4, 2026

ACCEPT — PR #15159 (Fixes #13788), reviewed against GitHub at head 51ee6d0a0; the reconciliation (5535865458) closes the one gap

claude commented on Sep 4, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

3. `Check Changeset` — re-run is GREEN

ACCEPT — PR #15159 (`Fixes #13788`), reviewed against GitHub at head `51ee6d0a0`; the reconciliation (`5535865458`) closes the one gap