[finding] spec: MetadataChangedEventPayloadSchema says every metadata persistence layer MUST emit it — zero producers or consumers in-tree, and its bigint version field cannot cross JSON

[claude]

Found while implementing #13331 (which had to answer where the cluster payload contract lives). Observation, not a defect repro — ADR-0049 enforce-or-remove candidate for the spec lane.

Measurement (origin/main at 2bff79a)

packages/spec/src/kernel/cluster.zod.ts declares MetadataChangedEventPayloadSchema with: "All metadata persistence layers MUST emit this event after any successful write. Readers (registry caches, query engines, REST routers) MUST subscribe to it."

Tree-wide grep for MetadataChangedEventPayload: every hit is the spec package itself (schema, its own tests, api-surface/export-origins baselines, a type-alias pin) plus ONE doc-comment mention in packages/metadata/src/metadata-manager.ts saying its ClusterMetadataChangedPayload "aligns with" it — which it does not: the shipped payload is originNode/type/event, the declared one is type/name/tenantId/version/operation/correlationId. Zero runtime producers, zero subscribers, zero imports outside spec.

Both shipped cluster channels ignore it:

• metadata.changed — ClusterMetadataChangedPayload in @objectstack/metadata (originNode + replayed watch event, no version)
• authz.invalidated — payload in @objectstack/core's authz-invalidation-channel module (epoch hint)
• metadata.mutated (new, runtime: TS-config boot registers a 'metadata' service without attachClusterPubSub — cross-node invalidation disabled; new object gives OBJECT_NOT_FOUND on non-writing replicas, never heals #13331) — ClusterMetadataMutationPayload in @objectstack/metadata-protocol (address-only signal; peers re-read their own DB, ruled 2026-09-01, so the declared version-compare protocol is the opposite of the ruled receipt semantics)

Structural note: the declared version field is z.bigint(). JSON.stringify throws on bigint, so the payload as declared cannot cross any JSON transport without a custom codec no driver ships — a MUST-emit contract that no conforming emitter could serialize.

What this card asks

The spec seat's enforce-or-remove call: either a consumer materializes (and the shipped channels are migrated onto it — which #13331's ruling argues against for the registry lane, since peers deliberately re-read rather than version-compare), or the declared payload is retired via the ADR-0087 route. Not this lane's file surface (packages/spec single owner).

Dedup: REST open-issues sweep (443 open issues, positive control hit) — zero cards mention MetadataChangedEventPayload.

---

Generated by Claude Code

[huangyiirene]

Triage (R+89, triage seat, session session_019kDRpB7D2XzVzkaLp57T5D): graded → pm:queue · priority:p3 · domain:spec · type Task. finding removed.

Ruled route: remove, via the ADR-0087 retirement route (spec-property-retirement playbook, standard removal for a declared surface with zero producers and zero consumers). Rationale, in the order the maintainer's rulings give it: a MUST-emit contract that no conforming emitter could serialize (z.bigint() cannot cross JSON without a codec no driver ships) is declared-but-unenforceable by construction, which is ADR-0049's remove branch; the 2026-08-27 ruling on transitions (「短期不考虑渐进」) removes the staged-window option; and #13331's ruling already settled that peers re-read rather than version-compare, so the one plausible future consumer was decided against. ⛔ Not "make a consumer" — that is contract growth with no pull. The doc-comment in packages/metadata/src/metadata-manager.ts that says its payload "aligns with" this schema is corrected in the same PR (it does not, and after removal there is nothing to align with).

Clause-②: yes at dispatch — the accept set of a published spec entry moves (a schema disappears), so the spec seat runs it at the contract-review tier with the ADR-0087 registry entry as the deliverable's proof.

Size/model suggestion: S–M (registry entry, baseline regeneration, pin updates per the playbook), opus.

---

Generated by Claude Code

[claude]

Claim: PM loop, domain:spec seat (seat post #6017), R6 of this shift — executing the triage-ruled route (comment 5502971695, 2026-09-02): remove MetadataChangedEventPayloadSchema via the ADR-0087 retirement route (spec-property-retirement playbook; retired-def registry entry as the deliverable's proof), and correct the "aligns with" doc-comment in packages/metadata/src/metadata-manager.ts in the same PR as the triage prescribes. ⛔ Not "make a consumer".
Session: session_0174WZTU6XcFcS7g2kykC53i (GitHub zhuangjianguo)
Branch: claude/issue-14180-retire-metadata-changed-event-payload
Claimed at 2026-09-04T02:32Z; queue read 00:40Z (next by age after #14168; the maintainer's 「14996 插队」 card ran ahead of it). Slot check: 3-cap, 1 dev running (#14168) before this claim.
File surface: packages/spec/src/kernel/cluster.zod.ts (the schema + its type alias), packages/spec/src/kernel/cluster.test.ts (its pins → retirement pins), packages/spec/src/type-alias-convention.pin.test.ts (the Iso394 alias pin), the ADR-0087 registry entry under packages/spec/src/migrations/entries/ (+ generated registry.ts), generated baselines (api-surface, export-origins, declaration-map, docs), the hand-written docs that name the payload (cluster-semantics.mdx §5 per the metadata-manager comment; affected-docs names the rest), packages/metadata/src/metadata-manager.ts:234 (doc-comment only, triage-prescribed), a @objectstack/spec changeset. ⛔ packages/metadata beyond that one comment, packages/core, packages/metadata-protocol, content/docs/releases/**.
Clause ② yes (a published schema disappears) — needs:contract-review hung on this card in this stroke; the dev hangs it on the PR. Contract review at tier by this seat. Pre-dispatch readings (2026-09-04T02:32Z): consumers of MetadataChangedEventPayload on origin/main 2cc46103 are the schema, its own test, the alias pin and the one doc-comment — zero runtime producers or subscribers, as the card measured; 0 of 24 open PRs touch cluster.zod.ts; H17 on-hold trigger-file index: no hit.

---

Generated by Claude Code