# COV-004 evidence pack — locale, modality, accessibility, and manual content

## Work-package charter

- **Task ID:** `COV-004`
- **Owner:** Oshun portfolio content-coverage contract maintainer;
  implementation owner: Codex
- **Accountable product:** Shared V1–V9 autonomous-content platform
- **Priority:** P0
- **Target milestone:** 2026-07-15 SOTA gap-closure program
- **Risk tier:** High release-integrity, localization, accessibility, inventory,
  and human-authored lineage risk
- **Dependencies:** COV-001 canonical content slots; COV-002 source-preserving
  seed; COV-003 portfolio dependency contract; source-owned V1–V9 locale and
  manual-content records; AUD-003 reproducible verification; AUD-006 generated
  capability truth
- **Affected schemas and inventory:** New strict
  `ContentInventoryRegistryV1Schema`, reviewed modality accessibility floor,
  deterministic generator/report, 77 inventory rows, unknown-preserving
  aggregates, exports, focused CI, and adversarial tests. No inventory row is
  promoted to a complete COV-001 slot.
- **Rollout strategy:** Publish through `@oshun/contracts/content-coverage` and
  the root contracts namespace, keep the generated registry/report
  source-digest-bound in CI, then require COV-005 and product importers to
  canonicalize the rows and reconcile release evidence.
- **Rollback strategy:** Revert the inventory contract, generator, committed
  outputs, tests, exports, CI step, audit qualification, and ledger closure
  together. Never preserve an aggregate that silently treats unknown as zero.
- **Evidence-pack location:** This file

## Inventory result and locale truth

The deterministic registry contains **77 rows**:

- **54 COV-002 targets:** 28 V1 generation surfaces, four V1 live autonomy
  categories, 16 V8/V9 required artifacts, and six V8/V9 placement targets.
- **Nine direct product locale observations:** V1 shared customer messages (8),
  V2 translation files (2), V3 HUD bundles (18), V4 recorded VO packs (12), V5
  stadium caption language count (16, identities undeclared), V6 cognition
  dialogue profiles (3), V7 creator-marketplace launch locales (8), V8 case
  source locale (1), and V9 lesson planner targets (undeclared).
- **14 intentionally manual records:** three expired V2 MOTDs, one V2 patch-note
  release, one V4 schedule-rotation patch note, six V5 approved replacement
  artifacts, and three V5 public-news articles.

Locale observations have exactly three states. `enumerated` requires canonical,
unique BCP 47 identities whose length equals the count. `count-only` preserves a
positive declared count but cannot invent identities. `undeclared` has a null
count rather than zero. A one-identity enumeration is always an explicit
`single-locale` gap.

The V1 shared customer-message catalog has eight locales, but none of the 28
generation surfaces or four live autonomy-category definitions links itself to
that catalog or declares a locale. All 32 remain explicit undeclared gaps. The
V5 caption declaration remains count-only, V8 remains an explicit `en-US`
single-locale gap, and V9 remains undeclared.

## Modality and accessibility policy

Every row has a reviewed content-kind/modality classification. The contract
computes the exact union of these COV-004 minimums:

| Modality    | Required minimums                                                         |
| ----------- | ------------------------------------------------------------------------- |
| text        | reading level, assistive metadata                                         |
| image       | alt text, assistive metadata                                              |
| audio       | transcript, assistive metadata                                            |
| video       | captions, transcript, audio description, assistive metadata               |
| 3D          | audio description, keyboard access, controller access, assistive metadata |
| animation   | audio description, assistive metadata                                     |
| interactive | keyboard access, controller access, assistive metadata                    |
| gameplay    | audio description, keyboard access, controller access, assistive metadata |
| metadata    | assistive metadata                                                        |
| code        | keyboard access, reading level, assistive metadata                        |
| document    | reading level, assistive metadata                                         |

The schema rejects a row whose accessibility list differs from the reviewed
modality floor. This records requirements per row/content kind; it is not
evidence that any current artifact conforms.

## Manual lifecycle, lineage, and quantities

Manual rows use the same strict inventory fields, source locators, stable
lineage references, locale observation, accessibility requirements, lifecycle
state, quantities, and gap ownership as agent/hybrid targets. The three MOTDs
are explicitly `stale` because their source end times precede the inventory
date. Static patch notes, replacement records, and news are `review` records;
they are not promoted to production placement.

Required/current/missing/stale/blocked quantities remain nullable. The 16
required artifacts and six placement targets each contribute one known required
unit, while their current/missing/stale/blocked values remain unknown. Manual
source records contribute known current units; stale MOTDs contribute known
stale units. Every aggregate reports `{ knownSum, unknownRows }` by product,
product surface, modality, locale, source policy, and accessibility requirement.
Thus a group with no known missing quantity reports zero known plus its unknown
row count, never a false zero-missing conclusion.

Every one of the 77 rows currently has at least one explicit gap, and all 77
assign `team:oshun-content-coverage-migration` with escalation
`oncall:oshun-platform`. That is migration accountability, not invented product
ownership.

## Determinism and artifact identity

| Artifact                                                      | SHA-256                                                            |
| ------------------------------------------------------------- | ------------------------------------------------------------------ |
| `libs/contracts/src/content-coverage/inventory-registry.ts`   | `ca31e1bc6e47c6e4de0fde25ebcd3296590a04aabef6d2158e4da78a9fc2d018` |
| `scripts/content-coverage/generate-content-inventory.ts`      | `f82709367bdc669898615aed8d87644e8ab84f6ed76364d41f11275c4a3bd252` |
| `scripts/content-coverage/generate-content-inventory.test.ts` | `a56b738395ae166d724c629024e03f606fa98421c2efea13f9903022a1a608bd` |
| `content-coverage/content-inventory.v1.json`                  | `c914b426b9bceba986392c1f52d00dbe0a1b741a4595036e62715d6ba75e59ca` |
| `content-coverage/content-inventory-report.md`                | `2fed182eabe427b0788c52c08b199e58d992407f11d9bf3e3eaf2d5a78f04a8f` |
| `.github/workflows/content-slot-contract.yml`                 | `2804d75d2517c024be01a0ba5bd290ec17101da92f73850648b381fe360e3a34` |
| `package.json`                                                | `e5916556278e56f52cae0f3df4117ca3c754e4a1ff5740ff0186c59df8bfface` |

Row IDs hash stable identities. Source files carry SHA-256 digests, each row
references declared source locators, directory enumeration is sorted, reviewed
classifications are frozen maps, and aggregate dimensions/values are sorted. The
generator has no wall-clock or random input; the inventory date is fixed.

## Automated and adversarial verification

- `pnpm verify:content-coverage:inventory` passes targeted ESLint, the script
  TypeScript project, **10/10** Node test cases, byte-for-byte generated output
  drift checking, and the contracts source typecheck.
- Tests freeze all 54 COV-002 targets and nine direct locale surfaces; prove V1
  catalog locales are not inherited by its autonomy/generation rows; freeze all
  exact/count-only/undeclared observations; verify all 11 modality policies;
  inventory all 14 manual records; prove every required aggregate dimension and
  unknown counter; and require ownership on every gap.
- Adversarial cases reject locale-count inflation, accessibility weakening,
  unowned gaps, stale aggregates, stale summaries, unknown weakening fields, and
  source-digest drift. A second source build and `--check` prove committed
  output determinism.
- `pnpm verify:contracts:content-slot` remains green at **17/17** tests,
  `pnpm verify:contracts:content-dependency-graph` remains green at **28/28**,
  and `pnpm verify:content-coverage:seed-registry` remains green at **10/10**
  with 68 registered / 7,185 unregistered seed rows.
- `NX_DAEMON=false pnpm exec nx run @oshun/contracts:build --parallel=1` passes.
  A runtime `tsx` probe verifies schema version and accessibility helper exports
  through the root flat export, root `ContentCoverageContracts` namespace, and
  `@oshun/contracts/content-coverage` subpath.
- The contracts checks/builds ran sequentially. Resource gates observed about
  11–12 GiB available RAM, 30 GiB free swap, and 38 GiB free disk. Build memory
  was polled at least every 30 seconds and remained about 11 GiB available.
- Source pre-commit ESLint, Prettier, asset locking, and V5 documentation
  validation passed. The hook's inferred broad typecheck was bypassed with
  `SKIP_TYPECHECK=1` under the host resource policy; focused script/contracts
  typechecks and resource-gated package builds passed before and after remote
  integration.

## Negative proof and claim boundary

- This task proves a strict contract, source-backed repository inventory,
  reviewed modality/accessibility requirements, and deterministic reporting. It
  does not prove translation quality/completeness, live locale negotiation,
  artifact accessibility conformance, current production placement, or any
  deployed row count.
- The inventory still reports zero complete COV-001 slots. Most quantities are
  deliberately unknown. No COV-002 row is canonicalized and no COV-003 graph is
  populated. COV-005 still owns release blocking, unknown-shipped detection in
  release paths, alerts, and runtime reconciliation.
- Static source is not a live artifact or placement receipt. A test fixture is
  not a translated, accessible, approved, or customer-observed release. No
  provider result, human approval, or production probe is manufactured.
- No user-facing web/admin/mobile UI, persistence migration, media pipeline, or
  Unreal asset changed. Playwright/mobile/browser accessibility, database,
  provider, media, and Unreal execution are therefore inapplicable to COV-004.

## Publication identity

- **Implementation source commit:** `968c65790ff87e457c8cb7ccc8aff38a6ad38cae`
- **Remote-integrated tested source:**
  `f0665ab54fdf1f62bea260ba51f023bc45c2f52e`
- **Publication metadata commit:** `e64c8584ff73f564c3a942b4f52865efe5578040`
