fix(pm): make dispatch-gates see the metadata TYPE registry's i18n coupling (#9219)

walkMetadataForms (packages/cli/src/utils/i18n-extract.ts) emits the
metadataForms i18n surface from two producers: METADATA_FORM_REGISTRY (the
17 *.form.ts modules, already covered by the form-module convention entry
in dispatch-gates.mjs) and DEFAULT_METADATA_TYPE_REGISTRY plus the
METADATA_FORM_REGISTRY module itself (packages/spec/src/kernel/
metadata-plugin.zod.ts, packages/spec/src/system/metadata-form-registry.ts).
Editing either of the latter two moves the same platform-objects bundles
PR #9113 had to regenerate, and derived nothing from dispatch-gates.

Both are exactly two known files, not a runtime-enumerated population, so
this closes the gap the same way SURFACE_MODULE and
check-type-check-coverage.mjs's ROOT_PROGRAM_COUPLED_SCRIPT already do:
two bare module-body path-literal constants declared directly in
check-i18n-bundles.mjs (the script check:i18n resolves to), read by the
ordinary path-literal derivation with no new CHANGE_KIND_GATES entry.
Pinned live in dispatch-gates.mjs's own self-test (existsSync + a matched
classifyEntry verdict for each), so a renamed or moved registry module
reddens that self-test instead of the derivation silently going blind on
this edge again.

Fixes #9144

Co-authored-by: Claude <noreply@anthropic.com>

Author: os-project-manager

scripts/check-i18n-bundles.mjs

@@ -115,6 +115,37 @@ import { findExtractConfigs, flagsFromDocstring } from './i18n-bundle-surface.mj
  */
 const SURFACE_MODULE = 'scripts/i18n-bundle-surface.mjs';
 
+/**
+ * The two producers of the `metadataForms` TYPE-LEVEL surface, declared as
+ * bare module-body path literals for the same reason as SURFACE_MODULE above
+ * (#9144). `walkMetadataForms` (packages/cli/src/utils/i18n-extract.ts) emits
+ * `metadataForms.<type>.label` / `.description` for every entry of
+ * `DEFAULT_METADATA_TYPE_REGISTRY` — including form-less types like
+ * `datasource`/`job`/`translation` — and separately reads
+ * `METADATA_FORM_REGISTRY` itself (the map, not the `*.form.ts` leaves it
+ * points at) to decide which types also get section/field labels. Editing
+ * either moves the same `platform-objects` bundles PR #9113 had to
+ * regenerate — and, unlike the `*.form.ts` leaves, neither carries a filename
+ * SURFACE_MODULE's own convention (`isMetadataFormModulePath`) can see.
+ *
+ * Not folded into SURFACE_MODULE: that module documents the ONE convention it
+ * enumerates at runtime (the `.form.ts` suffix), and these two files
+ * deliberately do not carry it — inventing a second filename convention for
+ * exactly two files would be the guess #9144 declined to make. A bare
+ * coupling constant was the option the card measured to have no downside
+ * beyond upkeep; see i18n-bundle-surface.mjs's header for the two rejected
+ * alternatives and why.
+ *
+ * This is per-coupling manual upkeep, deliberately, and it does not go quiet:
+ * dispatch-gates' own self-test pins that a card editing either path derives
+ * check:i18n, against the real files (existsSync) — delete either constant
+ * and that self-test reddens instead of the silence coming back. If either
+ * module is renamed or the registries merge, update the self-test's pins in
+ * the same change: the evidence goes with the claim, never ahead of it.
+ */
+const METADATA_TYPE_REGISTRY_MODULE = 'packages/spec/src/kernel/metadata-plugin.zod.ts';
+const METADATA_FORM_REGISTRY_MODULE = 'packages/spec/src/system/metadata-form-registry.ts';
+
 /** The one command this gate invokes per package, as oclif topic/command parts. */
 const EXTRACT_COMMAND_ID = ['i18n', 'extract'];
 const write = process.argv.includes('--write');

scripts/i18n-bundle-surface.mjs

@@ -50,17 +50,23 @@
  * form — the convention and the population coincide, with nothing left over on
  * either side.
  *
- * What it deliberately does NOT cover, measured and stated so the next reader
- * does not mistake silence for coverage: the type-level half of the same
- * surface. `walkMetadataForms` in packages/cli/src/utils/i18n-extract.ts emits
+ * What it still deliberately does NOT cover, stated so the next reader does
+ * not mistake silence for coverage: the type-level half of the same surface.
+ * `walkMetadataForms` in packages/cli/src/utils/i18n-extract.ts emits
  * `metadataForms.TYPE.label`/`.description` for every entry of
  * DEFAULT_METADATA_TYPE_REGISTRY (packages/spec/src/kernel/metadata-plugin.zod.ts),
- * and the registry module itself decides which forms are walked. Editing either
- * moves the same four bundles and matches no convention here, because neither
- * carries a filename that distinguishes it. Closing that edge needs an anchor
- * this module does not have, and the candidates trade off against each other
- * rather than being one obvious shape, so it is filed rather than guessed at:
- * issue 9144.
+ * and the registry module itself (packages/spec/src/system/metadata-form-
+ * registry.ts) decides which forms are walked. Editing either moves the same
+ * four bundles, and neither carries a filename this module's ONE convention
+ * (the `.form.ts` suffix) can see — inventing a second convention for exactly
+ * two files would be a guess, not a discovery. #9144 closed that edge WITHOUT
+ * extending this module: the two module paths are declared as bare coupling
+ * constants directly in check-i18n-bundles.mjs (METADATA_TYPE_REGISTRY_MODULE
+ * / METADATA_FORM_REGISTRY_MODULE, next to SURFACE_MODULE), which the
+ * derivation reads straight off that gate's own module body — no population
+ * to enumerate, since it is exactly two known files. See that pair's doc
+ * comment for why here was the wrong home and dispatch-gates.mjs's self-test
+ * for how the constants are kept honest.
  */
 
 import { readFileSync } from 'node:fs';

scripts/pm/dispatch-gates.mjs

@@ -1159,15 +1159,16 @@ export function reachesMetadataFormModule(path, modulePaths) {
  *     output for every card in the tree;
  *   - `check:i18n` walks `packages/` at runtime for files NAMED
  *     `i18n-extract.config.ts` and re-extracts each owning package's bundles.
- *     Its source is worse than silent: the path-ish literals it does carry are
- *     its CLI prerequisite and stale-dist checks (`packages/cli/dist/commands/
- *     i18n/extract.js`, `packages/spec/dist`, measured — eleven hints, none of
- *     them the population). So it matches nothing AND, having hints, never
- *     reaches the "undetermined" bucket either: before this entry existed, an
- *     edit to `packages/services/service-messaging/src/objects/` — which
- *     regenerates that package's four bundles — printed the gate in NEITHER
- *     half of the output. A gate the derivation cannot mention at all is the
- *     one shape this script must not produce; it cost a PR a CI round.
+ *     Its source names only three hints (measured, post-#9144): the shared
+ *     walk module (SURFACE_MODULE) and the two metadata-registry coupling
+ *     constants below — none of them the OWNING-PACKAGE population this entry
+ *     answers for. So it still matches nothing on an ordinary object/field
+ *     edit AND, having hints, never reaches the "undetermined" bucket either:
+ *     before this entry existed, an edit to
+ *     `packages/services/service-messaging/src/objects/` — which regenerates
+ *     that package's four bundles — printed the gate in NEITHER half of the
+ *     output. A gate the derivation cannot mention at all is the one shape
+ *     this script must not produce; it cost a PR a CI round.
  *
  * No per-card gate list derived from paths can ever name these, however the
  * derivation improves.
@@ -1251,6 +1252,32 @@ export function reachesMetadataFormModule(path, modulePaths) {
  * baseline. The day the last one opts out, no form module can move a committed
  * bundle and this entry stops firing on its own.
  *
+ * ## Why there is no THIRD i18n entry, for the type-registry edge (#9144)
+ *
+ * `walkMetadataForms` has a second edge the SECOND entry above does not reach:
+ * `DEFAULT_METADATA_TYPE_REGISTRY` (packages/spec/src/kernel/metadata-plugin.
+ * zod.ts) supplies `metadataForms.<type>.label`/`.description` for EVERY
+ * registry entry, including form-less types, and `METADATA_FORM_REGISTRY`
+ * itself (packages/spec/src/system/metadata-form-registry.ts, the map, not
+ * the `*.form.ts` leaves it points at) decides which types get section/field
+ * labels at all. Editing either moves the same bundles PR #9113 paid for —
+ * but unlike the `.form.ts` leaves, neither file carries a filename the
+ * `.form.ts` convention (or any convention) distinguishes, so a KIND entry
+ * here would need to invent one for exactly two files.
+ *
+ * That is not the same shape as the two entries above: this is not a
+ * runtime-enumerated population at all, it is two SPECIFIC, KNOWN files —
+ * the shape `SURFACE_MODULE` and `check-type-check-coverage.mjs`'s
+ * `ROOT_PROGRAM_COUPLED_SCRIPT` already use. So it is closed there instead:
+ * `check-i18n-bundles.mjs` declares both paths as bare module-body coupling
+ * constants (`METADATA_TYPE_REGISTRY_MODULE` / `METADATA_FORM_REGISTRY_
+ * MODULE`), which the ORDINARY path-literal derivation now reads directly off
+ * that gate's own source — no `CHANGE_KIND_GATES` entry, no `matches`
+ * function, nothing here to keep in sync. See that pair's doc comment in
+ * check-i18n-bundles.mjs for the full reasoning, and this file's own
+ * self-test for the live pins that keep the constants honest as the coupling
+ * they are: manual, per-file, and silently rottable if nothing watched it.
+ *
  * ## How these entries stay honest
  *
  * - Every `name` here is resolved against the families actually discovered in
@@ -2359,6 +2386,32 @@ function selfTest() {
   // than a pair of matching strings.
   t('the declared shared module exists', existsSync(join(ROOT, SHARED)));
 
+  // The same shape again, for the TYPE-registry edge of walkMetadataForms
+  // (#9144) — two specific, known files rather than a runtime-enumerated
+  // population, so they are closed as coupling constants in
+  // check-i18n-bundles.mjs rather than a third CHANGE_KIND_GATES entry. Both
+  // directions pinned LIVE: delete either constant and this reddens instead
+  // of the derivation going silently blind on that edge again.
+  const TYPE_REGISTRY = 'packages/spec/src/kernel/metadata-plugin.zod.ts';
+  const FORM_REGISTRY = 'packages/spec/src/system/metadata-form-registry.ts';
+  const i18nGateHints = readHints('scripts/check-i18n-bundles.mjs');
+  t('the i18n gate declares the type-level metadata registry module', covers(i18nGateHints, TYPE_REGISTRY));
+  t('the i18n gate declares the form registry module too (not just its *.form.ts leaves)', covers(i18nGateHints, FORM_REGISTRY));
+  const typeRegistryVerdict = classifyEntry({ files: ['scripts/check-i18n-bundles.mjs'], hints: i18nGateHints }, [TYPE_REGISTRY]);
+  const formRegistryVerdict = classifyEntry({ files: ['scripts/check-i18n-bundles.mjs'], hints: i18nGateHints }, [FORM_REGISTRY]);
+  t(
+    'so a card editing the type registry is MATCHED through that constant, not dropped as silent',
+    typeRegistryVerdict.verdict === 'matched' && typeRegistryVerdict.hits[0]?.hint === TYPE_REGISTRY,
+  );
+  t(
+    'and a card editing the form registry module is MATCHED through its own constant',
+    formRegistryVerdict.verdict === 'matched' && formRegistryVerdict.hits[0]?.hint === FORM_REGISTRY,
+  );
+  // Both declared paths are real files, so the four claims above are live
+  // rather than a pair of matching strings.
+  t('the declared type registry module exists', existsSync(join(ROOT, TYPE_REGISTRY)));
+  t('the declared form registry module exists', existsSync(join(ROOT, FORM_REGISTRY)));
+
   // ── A family's OWN script files as match keys (#8509) ─────────────────────
   //
   // Both directions are the product, and both are pinned: a card editing a