docs(adr): record the paper metadata-customization module's retirement in ADR-0005

[claude]

Fixes #13185

ADR-0005 named packages/spec/src/kernel/metadata-customization.zod.ts in two places as a kept, deliberately-retained model held for a future phase. That module has been deleted whole, so both sentences pointed at a file that no longer exists and promised a future phase ADR-0126 §6 has since ruled out.

Decides nothing new. This is a factual text correction to an accepted record, not a re-litigation of ADR-0005. Its own decision — the org-scoped overlay (allowOrgOverride on DEFAULT_METADATA_TYPE_REGISTRY, sys_metadata org rows, the layered code/overlay/effective read) — is untouched by the retirement and is neither edited nor re-summarized here.

Evidence chain for the correction

• Maintainer ruling of 2026-08-29 on issue [finding] spec ships a three-layer metadata-customization protocol (MetadataOverlaySchema, field-level change tracking) with zero consumers — published reference docs describe an unimplemented customization architecture #12057 (comment 5459699497), verbatim 「同意」, adopting retirement of the paper metadata-customization protocol under ADR-0049 enforce-or-remove.
• Executed by PR Retire the paper metadata-customization protocol with its full coupling set #13186, merged 2026-08-29T14:16Z.
• Superseding model: ADR-0126 §6 — wall 3 (upgrades rewrite the packaged base; the customer's choices live in the activation ledger and no upgrade merges into them) rules out the 3-way-merge future phase, and wall 4 supersedes the paper protocol outright ("nothing may build against it").

Verified on origin/main 7404925 before any edit: git cat-file -e origin/main:packages/spec/src/kernel/metadata-customization.zod.ts reports the path does not exist, while both stale lines were still present.

What changed — one file, two sites, one home for the record

Per Prime Directive 13, both original sentences are left standing as written; the correction is appended.

• Design principles (binding), item 3 — the substantive claim, so this is where the full record goes: an inline ⚠️ Correction (#13185, 2026-08-29) note giving the retirement, the evidence chain and the superseding model, and stating explicitly that the principle's own decision is unaffected (customizations are still full-JSON deltas, field-level patches are still not implemented). What changed is only the reserve the sentence pointed at: the RFC 7396 / 3-way-merge direction is closed rather than deferred.
• References bullet — a restatement of that same claim, so it gets the terse corrected fact (the file no longer exists, and what deleted it) plus a pointer to the one home above, rather than a second copy of the reasoning.

Convention followed

The inline ⚠️ Correction (#issue, date) shape, appended to the claim it corrects with the original wording left standing — the ADR-0088 precedent, applied to ADR-0126 §6.4 by PR #13139 earlier the same day. The split between the two sites here mirrors how that PR handled its own §1.1 restatement: a pointer to one home for the correction, not two copies. A strikethrough annotation on the References bullet was considered and rejected — striking the original text through re-reads the record rather than leaving it standing, which is the half of Prime Directive 13 this convention exists to satisfy.

Verification

Gate family derived from the tree, not recalled: node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack (re-derived after the commit; unchanged). All five run green at e1415210, quoting each gate's own verdict line:

• pnpm check:adr-anchors — ✓ check-adr-anchors --self-test: 106 assertions over the real auditAdrDirectory() / auditCitedNumbers() / assembleAnchors() paths.
• pnpm check:doc-authoring — ✓ doc authoring guard: 393 files clean — no bare metadata literals.
• pnpm --filter @objectstack/lint run check:doc-formula-expressions — ✓ check:doc-formula-expressions: 22 record-scoped formula example(s) across 426 files / 1451 TS blocks judged clean by @objectstack/formula.
• pnpm check:pm-governed-merges — ✓ check-governed-merges --self-test: 206 assertions ...
• pnpm check:adr-links — ✅ check-adr-links: 576 relative link destination(s) under docs/adr/ resolve

Two notes on that run:

• check:doc-formula-expressions was not on the dispatch's expected gate list; the path derivation added it. Its first two invocations exited 1 with PREREQUISITE NOT MET (@objectstack/formula, then @objectstack/lint, not built) — the gate's own text says "Nothing was measured", so those are not failures. After turbo run build for each, it measures clean.
• The links gate result is a real measurement of the new cross-link, not an assumption: the same gate run against the base tree 74049254 in a throwaway compare worktree reports 575 destinations, against 576 here, so the census delta is exactly the one relative link this PR adds (./0126-packaged-metadata-customization-model.md). No relative link is written to the deleted module — it stays a code span.

No changeset

docs/adr/** only; this publishes nothing from any package, so the diff carries no changeset and the PR is labelled skip-changeset — the same treatment PR #13139 had for its ADR-only diff.

Governed surface

docs/adr/** is a governed surface under Prime Directive 14. This PR stays draft: no AI seat marks it ready, arms auto-merge, or enqueues it. Human merge or pinned maintainer approval is the landing path, and the dispatching PM seat owns the terminal steps after review.

---

Generated by Claude Code