docs(changelog): correct the two published Seed.locale claims PR #17013 falsified, with erratum lines

[os-bill]

Fixes #17026

Clause-②: no — the landing surface is prose inside two already-compiled packages/*/CHANGELOG.md entries plus erratum lines. No schema key, no accepted-set change, no new exported symbol, no new key on a published payload. Diff is markdown only; no path under packages/**/src/** moves.

The ruling this PR implements, verbatim

Director seat, decision batch #119 item 1, 2026-09-12T04:15Z. Maintainer's own word to the batch presented as 1A·2A·3A·4D·5A: 「同意」.

A: correct the two published CHANGELOG entries in the repository, with a post-release erratum note.

1. packages/spec/CHANGELOG.md and packages/metadata-protocol/CHANGELOG.md: the two Seed.locale sentences ("no first-party call site supplies one", "the liveness ledger records seed.locale as experimental") are corrected to the post-feat(runtime, spec): AppPlugin supplies SeedLoaderConfig.locale, so the Seed.locale axis takes effect on the default boot path #17013 truth, each followed by one erratum line naming PR feat(runtime, spec): AppPlugin supplies SeedLoaderConfig.locale, so the Seed.locale axis takes effect on the default boot path #17013 and the correction date.
2. Before editing, run the release tooling's dry-run (changeset version on a scratch branch) and record that a manual edit inside an already-compiled section survives — the triage seat's open confidence gap; if it does not survive, stop and report.
3. skip-changeset: no package publishes anything from this change.

And the boundary that makes A legitimate, also verbatim:

Immutability belongs to the shipped npm tarball, which this ruling does not and cannot touch; it does not belong to the repository's markdown.

Step 2 first — the gate, run before any edit

Run on a local scratch branch off the same base, then discarded. Marker inserted into the already-compiled ## 17.4.0 section of both files (into the very paragraph this PR edits), then pnpm changeset version.

reading	value
marker count before edit	spec 0, metadata-protocol 0
marker count after edit, on disk	spec 1, metadata-protocol 1 (git diff --stat = 2 files, 2 insertions / 2 deletions)
pnpm changeset version	VERDICT changeset-version-exit 0 — "All files have been updated"
marker count AFTER changeset version	spec 1, metadata-protocol 1 — the manual edit SURVIVES
where the marker ended up	spec 2451 → 5654; metadata-protocol 128 → 898 — pushed down, content intact
lit control — did the tool actually rewrite the files?	yes: a new ## 17.5.0 section is prepended at the top, packages/spec/package.json moves 17.4.0 → 17.5.0, and the 250 pending changesets drop to 0
restore	git reset --hard + git clean -fd; git diff HEAD empty, 0 marker residue, 250 changesets back, scratch branch deleted

⇒ changeset version prepends and never rewrites an already-compiled section. The triage seat's own stated blind spot — "我没有验证过手工编辑 CHANGELOG 会不会被下一次发版覆盖或冲突 …… 但我没有测" — is now measured, and it does not block A.

Step 1 — what changed, and why each corrected sentence is true of today's tree

Both files carry the same compiled ## 17.4.0 entry for commit c1d8f98, byte-identical before and after this change. The single paragraph at packages/spec/CHANGELOG.md:2451 / packages/metadata-protocol/CHANGELOG.md:128 becomes four: two corrected paragraphs, each followed by one erratum line naming PR #17013 and the correction date 2026-09-12.

Claim 1 — "no first-party call site supplies one". Verified false on this tree, not taken from the card:

• packages/runtime/src/app-plugin.ts:1717 declares private resolveSeedLocale(), reading i18n.defaultLocale off the bundle.
• It is threaded into all three SeedLoaderRequests that plugin builds: lines 1349, 1434, 1791 (...(seedLocale ? { locale: seedLocale } : {})).

⇒ the corrected sentence says a first-party call site does supply config.locale and names the producer and the three sites.

Claim 2 — "the liveness ledger records seed.locale as experimental". Verified false: packages/spec/liveness/seed.json → props.locale.status reads live, with producer naming packages/runtime/src/app-plugin.ts#resolveSeedLocale.

One adjacent clause corrected in the same stroke, named rather than smuggled. "on the default boot path the axis is inert" sits in the same paragraph and is falsified by the same landing. Leaving it would have left a known-false clause beside a corrected one, which defeats the ruling's own purpose; it is corrected and called out here.

The corrected text does not overclaim. The ledger row's live is explicitly bounded, and the correction carries that bound rather than dropping it. Measured on this tree: three request builders outside AppPlugin still build a config with no locale key —

• packages/runtime/src/domains/packages.ts:1776 applyPublishedSeeds (package apply) — config: { defaultMode, multiPass, organizationId? }, zero occurrences of locale in the whole file;
• packages/metadata-protocol/src/protocol.ts:17490 applySeedBodies (draft publish) — same three keys;
• packages/cloud-connection/src/marketplace-install-local-plugin.ts:1510 runInlineSeed (marketplace install) — same three keys.

And SeedLoaderService#warnOnUnresolvedLocaleScope still exists (packages/metadata-protocol/src/seed-loader.ts:2592, called at :554), so the corrected paragraph keeps the sentence about the warning instead of deleting it.

House convention followed rather than invented

The dispatch pointed at #16859 as the precedent. ⚠️ Read to the end, #16859 did not in fact edit a published CHANGELOG — its own report states packages/spec/CHANGELOG.md "was not opened for writing" and that correction 1 was "confirmed foreclosed and left alone". So it establishes no erratum wording.

The convention that does exist was found by measurement — three entries already corrected in place in this repo:

• packages/spec/CHANGELOG.md:5257 — "As published, this sentence justified that by … that was false when written and is false now …" closing (Corrected after publication, #15058.)
• packages/spec/CHANGELOG.md:7401 and packages/lint/CHANGELOG.md:1214 — same shape, closing (Corrected after publication, #15026.)

⇒ this PR follows that: the published sentence is quoted in place and marked false, the correction follows in the same entry, and the last erratum line closes (Corrected after publication, #17026.) naming the card, as those three name theirs.

It is also exactly what the packages/*/CHANGELOG.md guardrail row now added to AGENTS.md by #16849 prescribes: "Factual error in a released entry → amend that entry in a dedicated docs-only PR, ⛔ never an erratum in a later entry and never a rider on code changes". This PR is docs-only, amends the entry itself, and adds nothing to a later entry.

skip-changeset — measured, not asserted

reading	value
diff file list vs merge base c744c0af332	exactly two paths, both CHANGELOG.md
any non-markdown path	none
.changeset/*.md added by this PR	0
packages/spec files[]	dist, json-schema, liveness, prompts, llms.txt, README.md, src/**/*.zod.ts, CHANGELOG.md, api-surface, spec-changes.json
packages/metadata-protocol files[]	dist, README.md, CHANGELOG.md

⚠️ Stated precisely so neither half misleads: CHANGELOG.md is in both packages' files[], so the corrected bytes do ship inside the next tarball either package happens to publish. What does not move is any package version, any API, any dist/, any source file, or any generated artifact — and because this PR adds no changeset, it schedules no version bump of its own. That is the sense in which "no package publishes anything from this change" holds.

Gates

Derived with node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack against the real change set, then reconciled with --ran.

• 54 derived families — 50 run green, 4 NOT MEASURED, 0 unrun. Reconciliation verdict, verbatim: ✓ dispatch-gates --ran: 54 derived famil(ies) accounted for — 50 run, 4 NOT-MEASURED (4 DERIVED from a recorded exit 3). Every exit code captured before any pipe.
• The 4 NOT MEASURED are check:dts-closure, check:dual-build-cjs-loads, check:lean-entry-closure, check:sourcemap-no-sources-content; all exit 3 = PREREQUISITE NOT MET, each refusing because there is no dist/ in this worktree ("Run pnpm build first. ⛔ This is NOT a pass: nothing was measured."). They read build artifacts only, which a markdown-only diff cannot move. Declared to CI, which builds first.
• Repo-wide pnpm lint (eslint . --no-inline-config) over the full population: exit 0 in 1m26s. No narrowing declared because none was needed. A targeted run over just these two files reports File ignored because no matching configuration was supplied. for both — markdown is outside eslint's configured population here.
• pnpm check:nul-bytes green, plus a manual control-character sweep of both touched files (grep -naP over the non-tab, non-newline C0 range plus DEL): no matches, with a printable lit control on the same files firing.
• Package test suites are not owed and were not run, stated rather than silently skipped: the diff compiles into nothing and is imported by nothing. The only test that mentions CHANGELOG.md as a file class is the action-owner-key-single-source pair, whose scanner is fixed to .ts sources.

Acceptance notes — out of scope, noted, not filed

• packages/core/src/plugin-contract.ts:69 carries a code comment asserting the opposite posture for a different entry: "⚠️ CHANGELOG.md is NOT a live document and is deliberately not corrected … supersedes them BY VERSION rather than by rewriting them." That predates the packages/*/CHANGELOG.md guardrail row gate/AGENTS: the Documentation Guardrails table has no row for packages/*/CHANGELOG.md — a released, consumer-shipped, generated file that is neither .changeset/ input nor content/docs/releases/ #16849 added to AGENTS.md, which now prescribes amending the entry. ⛔ Not touched by this PR and not filed — it belongs to whoever next revisits that entry, and the governing text is already the AGENTS.md row.

Not in play

⛔ content/docs/releases/ is untouched by this PR and by all of the ruling's options. This is packages/*/CHANGELOG.md, a different surface. Named here only so the red line is not read into a task it does not cover.

Authored by the domain:spec execution lane, session session_01MkQhmuuJAVDjmeWNixwDDH, on branch claude/issue-17026-seed-locale-changelog-erratum.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

⚠️ 2 changed file(s) yielded no anchor (packages/metadata-protocol/CHANGELOG.md, packages/spec/CHANGELOG.md), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files. Nothing else in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 2 changed package(s)).

What this run could not see

• 2 changed file(s) yielded no anchor (packages/metadata-protocol/CHANGELOG.md, packages/spec/CHANGELOG.md) — pages documenting those are invisible to this run
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.

Coarse fallback — 137 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json a61ae59f93302c1256330be8a6282f7be7f4a120 → packageMentionDocs.