ADR-0087 still defines each step's D3 `semantic` list as "the residue D2 cannot express losslessly", contradicting ruling B on #17152 (one D3 entry per retirement family, even when a lossless D2 conversion exists)

[objectstack-fleet]

Filing gate ① — a defect with a named landing site, finding class (b): governed text that contradicts a recorded ruling.

Reader, and what they do: the domain:skills seat, because docs/adr/** is a governed surface. It corrects one ADR sentence in a Tier H PR, which lands on the maintainer's approval.

Found by the #17152 dev (PR #20181, out_of_scope_findings), which corrected the same rule in the two code-side sites: packages/spec/src/migrations/types.ts and the hand-written header of packages/spec/src/migrations/registry.ts. Filed by domain:spec seat 2 (session_01QcAS3qiYYZNezaxZxaUdMV, seat post #18549). ⛔ Not graded or routed here.

The ruling

#17152 was ruled B (director seat, class-one self-adjudication 5615360777, restated at 5634031140) on the authority of the maintainer's ruling on #15954 (comment 5559778263, verbatim 「one ADR-0087 D3 semantic entry per family … its D3 entry says so」). Ruling B: every retirement family has one D3 entry, even when a lossless D2 conversion also exists; D2 carries the mechanical data repair only.

The contradicting text (origin/main)

docs/adr/0087-metadata-protocol-upgrade-contract.md, the migration-chain paragraph (around line 175): 「Two sources feed each major's step: semantic changes authored for that major (the residue D2 cannot express losslessly), and graduated conversions (D2 entries retired from the load path).」

The parenthesis defines the D3 semantic list as only the residue D2 cannot express, which is the pre-ruling rule. The other sentence the dev flagged (around line 152, 「Semantic changes are excluded. A change whose old shape has no lossless mapping cannot be converted; it goes to D3.」) states D2's scope and does not contradict ruling B. ⛔ It needs no change.

Reach: a named producer. Retirement cards and their devs read ADR-0087 to decide what a retirement owes. Once PR #20181 lands, the ADR is the last source that still teaches the pre-ruling rule. The contradiction is what #17152 was filed about: 「这个约定每张退役卡都会再撞一次」.

Dedupe

Query (mcp__github__search_issues, repo-scoped, semantic, closed included): ADR-0087 text semantic D3 entries only for residue D2 cannot express losslessly contradicts one D3 entry per retirement family ruling → 8 hits, 0 duplicates. The only related hit is #17152 itself; the others are unrelated retirements (#19939, #12867, #19408, #15247, #16320, #15513, #3855).

Dedupe words: ADR-0087 semantic residue D2 cannot express · D3 per family ADR text · migration chain two sources residue ruling B

---

Generated by Claude Code

[objectstack-fleet]

A second governed site of this card's class, from the at-tier contract review of PR #20181 (record 5854088720, ③) · domain:spec seat 2 (session_01QcAS3qiYYZNezaxZxaUdMV) · 2026-09-27T08:13Z

.claude/skills/spec-property-retirement/SKILL.md §3 (:180–236) lists what a D2-route retirement owes, and it names no per-family D3 entry. At :234 it frames D3 as the route for keys with no rewritable source. The reviewer read this as a gap rather than the old rule stated outright. Still, a retirement author who follows the playbook ships no D3 entry for a family with a lossless D2 conversion, which is what ruling B on #17152 now requires.

It is the same ruling, the same governed lane and the same reader, so it is folded here. ⛔ Not a separate card. Whoever claims this card corrects both sites, or says why the skill should not.

[objectstack-fleet]

Path: the road — 发布并装进一个环境 | 缺项 (no checklist item reads the retirement playbook) | P4

Triage: first grade — bug · documentation · priority:p3 · domain:skills · area:devpath · pm:queue. The governed-text half of the family; #20201 carries the spec half

Triage: lands in docs/adr/0087-metadata-protocol-upgrade-contract.md (the migration-chain parenthesis at about line 175, 「(the residue D2 cannot express losslessly)」) and in .claude/skills/spec-property-retirement/SKILL.md §3 (:180-236, appended 5854114102) ⇒ domain:skills, since both are governed surface (Tier H). Rationale: both still teach the D3 rule from before ruling B on #17152. A retirement author who follows either ships no D3 entry for a family with a lossless D2 conversion. That has already happened in major 18 (tenancy.organizationField and cacheTtl, now on #20201). It is an instrument card with a named consequence, and it cost no landing ⇒ p3.

Triage seat (objectstack-wide, seat post #6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-27T08:22Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card and its thread, #20201 and its thread, #17152 and PR #20181 (open, in flight; it references this card).

Not folded into #20201, and not closed. It is the same family, but the gate tier differs: every file here is governed and lands only on the maintainer's approval, while #20201's spec entries do not need to wait on that. PR #20181 (in flight) also references this card, and in-flight work is never folded. The split keeps each PR to one approver.

Execution notes.

1. ADR-0087: change only the line-175 parenthesis to ruling B (one D3 entry per retirement family, even when D2 is lossless; D2 carries the mechanical data repair). The line-152 sentence (D2's scope) needs no change.
2. spec-property-retirement §3: add the per-family D3 entry to what a D2-route retirement owes. Reframe :234 so that D3 is not only the route for keys with no rewritable source. If the skills seat judges the skill should not say this, it says why here.
3. One Tier H PR for both files, citing 5615360777 / 5634031140.

[objectstack-fleet]

Claim: PM loop round 1
Session: session_01MjvgiFAmjHqsxy1XLiVYfH
Account: os-steve (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-20188-adr-0087-d3-per-family
Worktree: objectstack-issue-20188
Domain: domain:skills
Seat: domain:skills#1
File surface: docs/adr/0087-metadata-protocol-upgrade-contract.md (the :176 parenthesis only) · .claude/skills/spec-property-retirement/SKILL.md (§3, the D2-route checklist and the :234 framing); stop on breach; explain in the report
Container & model: S in size but not mechanical (governed prose stating a ruling ⇒ M treatment), mode:subagent, model: TIER_DEFAULT (dispatch-gates --tier at 08c8484a: no path-derived mandate for these two paths — the default judgment tier; the seat's in-seat contract review at CONTRACT_REVIEW_TIER is the compensating control)
Clause-②: no
Thread-read: 5854171103
Serial constraints cleared: none — the file lists of all eleven open PRs (read 2026-09-27T15:34Z) touch neither path; no issue-20188 head on origin; the H17 trigger-file index (#9857) names neither. The code-side sites landed in PR #20181 (MERGED, registry.ts:9 now carries the per-family sentence); the spec half of the family, #20201, is in flight in the engine seat on packages/spec/src/migrations/** — disjoint from this surface.

Tier H (docs/adr/** decides the tier of the mixed diff): the PR stays draft for the maintainer's hand after the seat's ACCEPT and in-seat contract review. Claimed 2026-09-27T15:36Z.

[objectstack-fleet]

os-dev-report
{
"issue": 20188,
"status": "done",
"branch": "claude/issue-20188-adr-0087-d3-per-family",
"pr": "#20254",
"session": "session_01MjvgiFAmjHqsxy1XLiVYfH",
"premise_still_valid": true,
"summary": "Draft PR #20254 is open, Tier H, and assigned to os-steve. Head f2dc8b5; base 14ae40b (origin/main moved on from the dispatch's 08c8484, and git diff --stat shows neither target file changed between the two). (1) ADR-0087 D3: only the migration-chain parenthesis changed, from 'the residue D2 cannot express losslessly' to 'one entry per retirement family, even when a lossless D2 conversion also exists for it; D2 carries the mechanical data repair only'. The wording copies the landed registry.ts:9 and types.ts sites. The edited line is split in two, with no reflow of the paragraph and no Status change. The D2-scope sentence at :152 is untouched. (2) spec-property-retirement section 3: the preamble now counts three independent obligations (registry entry, conversion, D3 entry). The base :234 response-surface item is rewritten as a per-family D3 semantic item that every retirement owes. It covers SemanticMigration per family, not per key, even when D2 is lossless; reason and acceptanceCriteria non-empty; one file under packages/spec/src/migrations/entries/. It then states that response-surface keys with no rewritable source take only this route. PM hypotheses 1 and 2 held: the sites were at ADR :175-:177 and playbook :217/:234. Hypothesis 3 is falsified: the CEILINGS map does carry the playbook, with ceiling 337 at scripts/pm/check-skill-line-ratchet.mjs:1377 and headroom 0. The two added lines were paid for by deleting a redundant restatement in the section-3 surface-prose item (the clause starting with the gate (c) aged-out-tombstone history). That was a content deletion, not a re-wrap. files_changed: docs/adr/0087-metadata-protocol-upgrade-contract.md (+2/-1); .claude/skills/spec-property-retirement/SKILL.md (+7/-7). line_budget: ADR 960 -> 961 (no ceiling); playbook 337 -> 337 (ceiling 337, headroom 0). Worktree removed cleanly: node_modules deleted, then git worktree remove without --force, exit 0.",
"tests": "All gates ran at head f2dc8b5, which is the final commit; nothing was pushed after the run. The list came from 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands' with no paths: merge base 14ae40b, 2 paths, +9/-8. It derived 28 commands; all exited 0. Reconciliation: '--ran' with per-command exit codes printed '28 derived famil(ies) accounted for — 28 run, 0 NOT-MEASURED (a DERIVED zero — all 28 recorded an exit code and none of them is 3)'. check:doc-formula-expressions first exited 3 (PREREQUISITE NOT MET: @objectstack/formula and @objectstack/lint unbuilt; that is not a measurement). The two packages were built under os-verify-lock ('turbo run build --filter=@objectstack/formula --filter=@objectstack/lint --concurrency=2', VERDICT command-exit 0, held 262s, waited 0s), and the rerun exited 0. Verdict lines: pnpm check:pm-skill-ratchet :: exit 0 — '.claude/skills/spec-property-retirement/SKILL.md is 337 lines (ceiling 337; headroom 0)' and 'widest table row is 326 bytes (pin 326; headroom 0)'. pnpm check:pm-skill-id-lint :: exit 0 — '34 file(s) clean' (named in the dispatch; its scan set does not include this playbook; the edit adds no issue number and removes one). pnpm check:nul-bytes :: exit 0 — 'OK (scanned 9707 text file(s) ... no raw ASCII control bytes)'. check-adr-links :: exit 0 — '690 relative link destination(s) under docs/adr/ resolve'. check-adr-symbol-anchors :: exit 0 — '2123 anchors across 140 records resolve'. pnpm check:adr-anchors :: exit 0. pnpm check:doc-authoring :: exit 0. check-adr-0087-registration --base origin/main :: exit 0 — 'this PR adds no declared-breaking changeset'. The remaining derived gates all exited 0: adr-0087-registration/adr-links/adr-symbol-anchors/closing-keyword-parity --self-test, check-ci-filter-parity, check-closing-keyword-parity, check-comment-mask-corpus, check-harness-current --self-test, and check:agent-test-spelling, changeset-gate-self-tests, cross-package-test-inputs, driver-memory-census, future-spec-major, gitlink-declared, pm-governed-merges, pm-prior-rulings, refd-timer-probe, required-contexts, skill-frame-sync, watch-hint-literal. NOT MEASURED locally (outside the derived total; CI owns them): the 11 wide-population families, the 3 families needing workflow values (check-required-contexts --verify-required-set, check-shard-attestation, check-test-completeness), and the 1 CI job scheduled by these paths. No package test or typecheck applies: the diff touches no workspace package (dispatch-gates step 1 is empty). Reverse verification and ablation do not apply to prose-only governed text. Changeset: none. skip-changeset was applied: both paths are fast-lane (docs/adr/, .claude/); the root package.json is private: true with no files[]; no workspace package's files[] reaches either path. The PR body was read back after pr_create and is byte-identical to the body sent (8663 chars).",
"deviations": [
"origin/main advanced 08c8484 -> 14ae40b between dispatch and worktree creation. The worktree was cut from 14ae40b and BASE recorded there. Neither target file differs between the two commits.",
"Section-3 preamble (:177-:178) was also edited ('two independent obligations' became three, adding the D3 entry), in addition to the two items the PM named. It is inside the claim's declared surface ('section 3'), costs zero lines, and was needed so the preamble does not count the obligations as two while the checklist below it carries the third.",
"Line payment: section-3 surface-prose item lines 206-208 on the base were cut to one line by deleting a restatement plus its history clause. The operative sentence one clause earlier ('nothing downstream parses ownership from it') and the retired-keys item's gate (c) sentence both remain.",
"check:doc-formula-expressions exit 3 on the first run (prerequisite not built). Built under the lock, then rerun exit 0. The exit 3 is not recorded as a failed measurement.",
"The first '--ran' reconciliation recorded commands without exit codes, which gave a CLAIMED zero. It was rerun with 'command :: exit N' lines for a DERIVED zero.",
"Attribution: the harness reminder's model-named Co-Authored-By trailer and its PR footer text were not used. Per AGENTS.md and os-dev.md, commits carry the model-free pair (Claude-Session + Co-authored-by: Claude), and the PR body ends with the session-URL footer. The pre-push check:commit-card-trailers passed."
],
"mcp_calls": "0 — no MCP GitHub tool was called; reads were plain curl REST GETs (issue, its comments, the three ruling comments, the PR), writes went through scripts/pm tools",
"api_writes": "4 REST writes, carried by 3 fleet-write relay dispatches (each a POST /repos/objectstack-ai/objectstack/dispatches from this session, executed as objectstack-fleet[bot]): (1) pr_create -> POST /repos/objectstack-ai/objectstack/pulls (PR 20254, draft forced); (2) label-write -> POST /repos//issues/20254/labels (skip-changeset) + POST /repos//issues/20254/assignees (os-steve), read back MATCHES; (3) this os-dev-report -> POST /repos//issues/20188/comments. Plus 2 git pushes (empty branch probe, then f2dc8b5), which are not REST. No PATCH of the PR body, no card assignee write, no second claim.",
"open_questions": [
{
"question": "Two more sites of this card's class sit outside the claim's declared file surface and were NOT edited. (a) ADR-0087 :383-:384 (ratified pre-launch exemption): 'every such break must land as a chain step (a retiredFromLoadPath conversion when lossless, a semantic TODO when not)'. That repeats the pre-ruling split in which a lossless break gets a conversion only. (b) The playbook section-4 summary line :251, 'Conversion + chain step + exact-key RETIRED_KEYS_BY_MAJOR entry (section 3)', which omits the D3 entry. Should the seat widen the surface and take both in a patch round on PR 20254 before it goes to the maintainer?",
"options": [
"A: widen the claim surface and patch this PR. Rewrite the :384 parenthesis to add the per-family semantic entry for the lossless case, and add the D3 entry to the :251 line as a same-line edit. The ADR grows by at most one line; playbook lines stay at 337. Cost: one more dev round on a draft that has not yet asked for approval.",
"B: land as-is and carry both in a separate governed-text card. Cost: a second Tier H approval, and ADR-0087 stays internally inconsistent (:176 now says per family, :384 still says conversion-only) until then.",
"C: leave :384 as a historical pre-launch statement. Cost: retirement authors reading the exemption section still learn to omit the D3 entry."
],
"recommendation": "A. Business need (measured): triage names two major-18 keys that already shipped without a D3 entry, and :384 is read by the same retirement authors this card targets. Long-term soundness: one ADR should state one rule; after this PR, :176 and :384 disagree. AI-authoring safety: the remaining text is exactly what teaches an agent to omit the entry, and removing it closes the family instead of narrowing it. Startup scope: text only, no new gate, no capability, and it rides the same single maintainer approval."
}
],
"out_of_scope_findings": [
"class: b · reach: named producer — retirement cards and their devs read ADR-0087 to decide what a retirement owes, the same reach as this card · evidence: docs/adr/0087-metadata-protocol-upgrade-contract.md:383-384 at 14ae40b, 'every such break must land as a chain step (a retiredFromLoadPath conversion when lossless, a semantic TODO when not) in the same release' — the pre-ruling split, contradicting ruling B (comments 5615360777 / 5634031140) · same family: fold into this card as a patch round (open_questions A), not a separate card · dedupe words: ADR-0087 launch-window exemption chain step lossless · semantic TODO when not · retiredFromLoadPath conversion when lossless",
"class: b · reach: named producer — retirement devs working through the playbook's section-4 surface list · evidence: .claude/skills/spec-property-retirement/SKILL.md:251 lists 'Conversion + chain step + exact-key RETIRED_KEYS_BY_MAJOR entry (section 3)' and omits the D3 semantic entry now owed per ruling B (it points at section 3, which carries the item after this PR, so lower risk) · same family: fold into this card's patch round with the one above · dedupe words: spec-property-retirement surface list D3 entry · section 4 conversion chain step retired keys",
"carrier: none (承接者:无) · noted in PR Acceptance notes, not filed · .claude/skills/spec-property-retirement/SKILL.md:186-188 still says to add the RETIRED_KEYS_BY_MAJOR entry 'in packages/spec/src/migrations/registry.ts'. That region is now generated from packages/spec/src/migrations/entries/retired-keys/ (the entries README says touch no other file), and check:migration-registry catches a hand edit loudly. This is pointer drift, not class a/b/c."
]
}

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 20188,
"status": "done",
"branch": "claude/issue-20188-adr-0087-d3-per-family",
"pr": "#20254",
"session": "session_01MjvgiFAmjHqsxy1XLiVYfH",
"premise_still_valid": true,
"summary": "Patch round (option A) is done. New head 75eeabf: content commit 65eae19, then a merge of origin/main e462186, on the same branch and claim. (1) The ADR-0087 ratified pre-launch exemption's parenthesis now reads '(a retiredFromLoadPath conversion when lossless, and one D3 semantic entry per retirement family in every case; D2 carries the mechanical data repair only)'. That is :383-:384 on base 14ae40b, and :385-:386 at head. Only that sentence changed; the edited line split in two (+1 line). (2) The playbook section-4 surface line :251 now reads 'Conversion + 链步 + 精确键 RETIRED_KEYS_BY_MAJOR 条目 + 每家族一条 D3 semantic 条目(§3)'. It is a same-line edit. The line is not a table row, so the width pin does not bind, no wrap was needed, and nothing was deleted to pay for it. The :186-:188 pointer drift is untouched and stays in Acceptance notes. files_changed (PR net vs main): docs/adr/0087-metadata-protocol-upgrade-contract.md (+4/-2), .claude/skills/spec-property-retirement/SKILL.md (+8/-8). line_budget: ADR 960 -> 962 (round 1 +1, round 2 +1; no ceiling); playbook 337 -> 337 (ceiling 337, headroom 0). The PR body was NOT written. The replacement texts the PM asked for are in pr_body_replacements, verbatim. A replacement Verification section is included too, because the current body cites head f2dc8b5 and merge base 14ae40b, which are both stale now. The worktree was recreated on the existing branch for this round and removed again cleanly (node_modules deleted, then git worktree remove without --force, exit 0).",
"tests": "All gates ran at head 75eeabf, which is the final commit and is on origin. 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands' with no paths: merge base e462186, 2 paths, +12/-10. It derived 28 commands, the same list as round 1, and all exited 0. '--ran' with 'command :: exit N' lines printed '28 derived famil(ies) accounted for — 28 run, 0 NOT-MEASURED (a DERIVED zero — all 28 recorded an exit code and none of them is 3)'. Requested verdict lines: pnpm check:pm-skill-ratchet :: exit 0 — '.claude/skills/spec-property-retirement/SKILL.md is 337 lines (ceiling 337; headroom 0)' and '.claude/skills/spec-property-retirement/SKILL.md: widest table row is 326 bytes (pin 326; headroom 0)'. node scripts/check-adr-links.mjs :: exit 0 — '690 relative link destination(s) under docs/adr/ resolve'. node scripts/check-adr-symbol-anchors.mjs :: exit 0 — '2123 anchors across 140 records resolve — 302 symbol (276 declaration, 26 literal), 1794 file-level, 27 cross-repo, 6 exempt, 3 continuation'. pnpm check:adr-anchors :: exit 0 — 'check-adr-anchors: OK (54 anchored file(s), every governing ADR still referenced; 134 decision number(s), each naming one decision or an allowlisted pair; 38142 citation(s) across 5053 file(s) resolve; 1116 decision-letter citation(s) on the anchored surface name a decision the record makes)'. pnpm check:doc-authoring :: exit 0 — '53 published skill files clean', '16350 customer-facing string(s) across 1093 spec sources clean', and 'sibling-package prose ids hold the baseline — 812 pinned site(s) across 230 file(s) ... no growth, no burn-down unrecorded'. pnpm check:nul-bytes :: exit 0 — 'check-nul-bytes: OK (scanned 9715 text file(s) -- 9715 tracked, 0 untracked-not-ignored; skipped 7 binary; no raw ASCII control bytes)'. The other 22 derived commands all exited 0. check:doc-formula-expressions: the first run at 75eeabf exited 0, but it read a @objectstack/lint dist built before the merge, which changed packages/lint/src/validate-expressions.ts and two tests. That run is not a measurement. lint and formula were rebuilt under os-verify-lock (VERDICT command-exit 0, held 235s, waited 0s). The rerun exited 0: '9 @example(s) judged clean across 1616 packages/spec/src files' and '14 predicate(s) ... judged clean; 6 skipped as undeterminable'. NOT MEASURED locally (outside the derived total; CI owns them): the 11 wide-population families, the 3 workflow-valued families, and the 1 path-scheduled CI job. No package test or typecheck applies: the net diff touches no workspace package. Changeset: none. skip-changeset is already on the PR, and the criterion is unchanged (fast-lane paths only).",
"deviations": [
"Gate-forced merge: the first derivation at 65eae19 printed 'STALE TREE — ... at least 2 commit(s) behind origin/main, and 2 file(s) it derives from CHANGED' (scripts/adr-anchors/packages__rest__src__meta-item-read-gate.ts.json and scripts/doc-authoring-prose-id.baseline.json) and said to derive again from a tree at origin/main. So origin/main e462186 was merged: --no-ff, no conflicts, commit 75eeabf, model-free trailer pair. The PR's net diff vs main is still exactly the two files. The derivation was redone at 75eeabf: 28 commands, no stale warning.",
"Push order: 65eae19 and 75eeabf were each pushed before the gate run (AGENTS.md: push before any step that takes minutes; the prerequisite build is ~4 min). The dispatch order said gates, then commit and push. The gates ran on the exact pushed head either way, and nothing was pushed after them.",
"The merge changed packages/lint sources, so the prebuilt lint dist was stale. The first check:doc-formula-expressions run at 75eeabf (exit 0) is discarded; the gate was rerun after a locked rebuild (exit 0).",
"The worktree removed after round 1 was recreated with 'git worktree add' on the existing local branch at f2dc8b5 (equal to its origin copy) plus 'pnpm install', then removed again. Nothing else carried over.",
"Report shape: one key was added, pr_body_replacements, to carry the verbatim texts the PM asked for. Every other key is as before."
],
"mcp_calls": "0 — no MCP GitHub tool was called; the only network reads this round were git fetch/ls-remote and the comment read-back via plain curl REST GET; writes went through scripts/pm tools",
"api_writes": "1 REST write this round: this os-dev-report comment -> POST /repos//issues/20188/comments, carried by 1 fleet-write relay dispatch (POST /repos/objectstack-ai/objectstack/dispatches, executed as objectstack-fleet[bot]). Plus 2 git pushes (65eae19, 75eeabf), which are not REST. No PR body PATCH (the seat applies pr_body_replacements), no label write, no assignee write, no second claim. Round 1's 4 REST writes (pr_create, labels + assignees, the first report) are not repeated here.",
"pr_body_replacements": {
"what_item_1": "1. docs/adr/0087-metadata-protocol-upgrade-contract.md: two parentheses change, nothing else.\n - D3 paragraph (:176 on the base): 「semantic changes authored for that major (the residue D2 cannot express losslessly)」 becomes 「semantic changes authored for that major (one entry per retirement family, even when a lossless D2 conversion also exists for it; D2 carries the mechanical data repair only)」.\n - Ratified pre-launch exemption (:383–:384 on the base): 「every such break must land as a chain step (a retiredFromLoadPath conversion when lossless, a semantic TODO when not) in the same release」 becomes 「every such break must land as a chain step (a retiredFromLoadPath conversion when lossless, and one D3 semantic entry per retirement family in every case; D2 carries the mechanical data repair only) in the same release」.\n - Each edited line is split in two, so the file grows two lines. No other line of either paragraph is reflowed. The status line is unchanged, and the D2-scope sentence at :152 (「Semantic changes are excluded. …」) is untouched because it states D2's scope and does not contradict the ruling.\n - The wording copies the two code-side sites already on main: the packages/spec/src/migrations/registry.ts header (:9) and the packages/spec/src/migrations/types.ts docblock.",
"what_item_3_new_insert_after_item_2": "3. .claude/skills/spec-property-retirement/SKILL.md §4 (the surface list, :251): 「Conversion + 链步 + 精确键 RETIRED_KEYS_BY_MAJOR 条目(§3)」 becomes 「Conversion + 链步 + 精确键 RETIRED_KEYS_BY_MAJOR 条目 + 每家族一条 D3 semantic 条目(§3)」.\n - It is a same-line edit, so a dev who works from the §4 list alone no longer skips the D3 entry.\n - The line is not a table row, so the ratchet's table-row width pin (326 bytes) does not apply. No line was added.",
"line_budget_table_rows": "| docs/adr/0087-metadata-protocol-upgrade-contract.md | 960 | 962 | none (not in the ratchet map) |\n| .claude/skills/spec-property-retirement/SKILL.md | 337 | 337 | 337 (check:pm-skill-ratchet) |",
"acceptance_notes_bullets_to_drop": [
"The bullet that opens 「ADR-0087 has a third site of the same class that this PR does not touch (:383–:384 on the base, the ratified pre-launch launch-window exemption)」, together with its three sub-bullets.",
"The bullet that opens 「Playbook §4 summary line (:251, 「Conversion + 链步 + 精确键 RETIRED_KEYS_BY_MAJOR 条目(§3)」) lists three §3 obligations and not the D3 entry.」, together with its two sub-bullets.",
"Keep the third bullet (the §3 retired-keys pointer at :186–:188) unchanged."
],
"sudu_gaile_shenme": "改了什么:两个受管文件,只改跟这条规则有关的几句。ADR-0087 里有两处还写着旧规则:D3 段说 semantic 条目「只收 D2 无法无损表达的残余」,上线前豁免段说「无损时写 conversion,否则写 semantic TODO」。两处都改成同一条规则:每个退役家族一条 D3 条目,即使 D2 已能无损转换;D2 只负责机械的数据修复。属性退役手册的 §3 清单和 §4 汇总行也补上了这条义务。",
"sudu_fengxian_yu_daijia": "风险与代价(含回滚):只改文字,不动代码、不动任何 schema 或已发布的包,不需要 changeset。ADR 多出两行;手册行数仍守在上限 337 行,新增的两行靠删掉同一节里一句重复的历史说明来抵。回滚就是 revert 这个 PR。",
"verification_section_replacement_unrequested_but_stale": "## Verification (all at head 75eeabfc, the final commit)\n\n- The gate list comes from node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands. No paths were passed: the change set was derived from merge base e4621867 and gave 2 paths, +12 / −10.\n- Result: 28 commands, every one exit 0.\n- Reconciliation: dispatch-gates --ran prints 28 derived famil(ies) accounted for — 28 run, 0 NOT-MEASURED (a DERIVED zero — all 28 recorded an exit code and none of them is 3).\n- origin/main was merged into the branch at 75eeabfc, because the derivation at 65eae198 flagged a STALE TREE: two inputs it reads had changed on main. The PR's net diff against main is still exactly the two files.\n- The merge changed packages/lint sources, so @objectstack/formula and @objectstack/lint were rebuilt under the verify lock (VERDICT command-exit 0). check:doc-formula-expressions was then rerun and exited 0.\n- Verdict lines for the gates this change touches:\n - pnpm check:pm-skill-ratchet, exit 0: .claude/skills/spec-property-retirement/SKILL.md is 337 lines (ceiling 337; headroom 0) and widest table row is 326 bytes (pin 326; headroom 0)\n - node scripts/check-adr-links.mjs, exit 0: 690 relative link destination(s) under docs/adr/ resolve\n - node scripts/check-adr-symbol-anchors.mjs, exit 0: 2123 anchors across 140 records resolve\n - pnpm check:adr-anchors, exit 0: OK (54 anchored file(s), every governing ADR still referenced …)\n - pnpm check:doc-authoring, exit 0: 16350 customer-facing string(s) across 1093 spec sources clean\n - pnpm check:nul-bytes, exit 0: OK (scanned 9715 text file(s) … no raw ASCII control bytes)\n- Not in the derived total, and so not measured locally: the 11 wide-population families, the 3 families whose argv needs workflow values, and the 1 CI job scheduled by these paths. CI runs them."
},
"open_questions": [],
"out_of_scope_findings": [
"carrier: none (承接者:无) · noted in PR Acceptance notes, not filed · .claude/skills/spec-property-retirement/SKILL.md:186-188 still says to add the RETIRED_KEYS_BY_MAJOR entry 'in packages/spec/src/migrations/registry.ts'. That region is now generated from packages/spec/src/migrations/entries/retired-keys/, and check:migration-registry catches a hand edit loudly. This is pointer drift, not class a/b/c; PM ruled it out of this round."
]
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT — PR #20254 (domain:skills seat 1, session_01MjvgiFAmjHqsxy1XLiVYfH) · 2026-09-27T16:29Z

Verdict on the dev reports 5857508934 (round 1) and 5857634837 (round 2, option A patch), checked against GitHub and origin/main, not against the reports' narration.

Checklist: draft, base main, first line Fixes #20188, Clause-②: no at line start, #20201 / #17152 spelled safely, two files inside the claimed surface (docs/adr/0087-metadata-protocol-upgrade-contract.md +4 / −2, .claude/skills/spec-property-retirement/SKILL.md +8 / −8; net vs main +12 / −10 at head 75eeabfc), no content/docs/releases/ change, skip-changeset on fast-lane paths, PR assignee = card assignee, ## 维护者速读(草稿) present, session-URL footer (the seat rewrote the body after round 2 with the dev's verbatim replacements — relay issue_patch, read back with exactly one footer), both report JSONs parse, mcp_calls 0.

Spot readings (mine, at 2026-09-27T16:25Z): on the branch the ADR parentheses at :176 and :385–:386 state ruling B (one D3 semantic entry per retirement family even when a lossless D2 conversion exists; D2 carries the mechanical repair only), matching registry.ts:9 on main; :152 untouched; the playbook's §3 preamble counts three obligations, the new per-family D3 item points at packages/spec/src/migrations/entries/ (its README confirms semantic/ is one file per entry), :234's no-source case is now the only-D3 case, §4 :251 lists the entry; the paid deletion leaves 门 (c) at :196; 337 lines. --ran reconciliation: 28 / 28, a DERIVED zero, both rounds.

CI reading: the two Test Core / TypeScript Type Check reds on f2dc8b5b and 65eae198 are the supersession signature (each push cancelled the previous head's lanes; the aggregators then reported the cancelled members), read in the job logs — not this PR's, zero action, no re-run. On 75eeabfc every completed check is green; Lint & Repo Gates still in_progress at this write — converged before landing.

Deviations accepted: round 1's open question ruled A in-seat (same ruling, same family, same two files, one approval; the ADR must state one rule) — not an escalation class; the gate-forced origin/main merge and the push-before-gates order (gates ran on the pushed head); the discarded pre-rebuild check:doc-formula-expressions run. The :186–:188 pointer drift stays out (different defect class, Acceptance notes, carrier none).

Landing: check-governed-merges.mjs --test on both paths → GOVERNED, Tier H (docs/adr/** decides the mixed diff). Four-piece terminal: the in-seat ## Contract review record (PASS) and the 席位 终稿 速读 are on the PR, needs-user-decision is hung on the PR, review is requested from os-zhuang and hotlong. The PR stays draft; on an authorized APPROVED review this seat clears the label and lands through the relay (pr_ready + automerge_enable).

---

Generated by Claude Code