# V7-003 bounded flagship region

Date: 2026-07-25

Bounded-region source commit: `a21a5593d90a3ff3f0a85d9bf6045ea2ea9cd30e`

Measured-revision source commit: `d8d564fa633c12e5fbbc7b23479676487d622a7b`

## Claim

`generateV7FlagshipProceduralRegion()` produces one deterministic, bounded
procedural evidence region: `region:mawu-market-quarter:v1`. A complete
regeneration yielded manifest digest
`32e93d6c7b7ef1cbbf208aa62f1a2a643bd2561bca54fc34986005a10809f89f` and passed
the independent manifest evaluator with no violations.

The generation class is explicitly `procedural-reference-baseline`. This is not
a product-model result, UE map, cook, live realm, playtest, human review,
signature, or canary.

## Governed generation inputs

The region composes the existing four-tool Mawu policy evaluation. Its
`text-to-3d`, `text-to-texture`, `text-to-world`, and `text-to-npc` outputs are
generated by the labeled reference procedural backend. Each source artifact must
be usable, semantically grounded, attributed, content-hash-bound, and record the
exact backend in provenance before the region evaluator accepts it. The manifest
retains all four artifact IDs and hashes.

The generator does not call the product gateway and does not inherit product
backend claims. It reuses the governed procedural outputs specifically as a
measurable baseline.

## Measured manifest

| Dimension                    |        Result |
| ---------------------------- | ------------: |
| Region bounds                | 128 m × 128 m |
| Terrain cells                |            16 |
| Governed source artifacts    |             4 |
| Placed assets                |             4 |
| Scheduled NPC roles          |             3 |
| Balanced ledger transactions |             5 |
| Dependency-checked quests    |             2 |
| Complete locales             |             3 |
| Required localization keys   |            15 |

Terrain covers the exact 4×4 grid with unique coordinates and cells. Asset
placements must reference a retained source artifact and fall inside the region.
NPCs cover guide, vendor, and steward roles, reference the governed NPC
artifact, and have valid bounded schedules.

The `MARKET_CREDIT` fixture ledger contains issuance, treasury, player, vendor,
and sink accounts. Every transaction has at least two integer postings to known
accounts and sums to zero. Both quests bind explicit existing target assets and
reward transactions; their dependency graph contains no missing nodes or cycles.

English, Spanish, and French contain every region, asset, NPC, schedule, quest,
objective, and caption key. The accessibility record requires captions,
non-color state markers, keyboard/gamepad parity, 200% text scaling, reduced
motion, and positional-audio indicators.

## Fail-closed checks

The evaluator rejects:

- a modified or malformed manifest digest;
- invalid or unbounded region dimensions and incomplete terrain coverage;
- missing, duplicated, ungrounded, unattributed, or non-reference source
  artifacts;
- missing/out-of-bounds assets or incomplete NPC roles/schedules;
- duplicated accounts/transactions or any non-zero-sum ledger transaction;
- duplicate, cyclic, missing-dependency, or reward-less quests;
- any missing locale string; and
- any accessibility-floor regression.

The adversarial specifications exercise ledger/reward corruption, incomplete
French localization, missing captions/reduced text scaling, and post-seal
tampering.

## Measured simulation and governed revision

The deterministic headless simulator replayed 32 sessions over the exact sealed
region. It derives reachability from the walkable terrain graph, navigation
success from reachable placed assets, quest completion from each quest's
structured target asset IDs, and ledger integrity from replayed zero-sum
postings.

The before report
`deb37938cb0f4ea5c4f8c9bf6e96c5f20910bfcc16da2dbd3e2e48ec79b5c31f` measured 128
navigation attempts, 96 successes, 64 quest attempts and completions, 160 ledger
transaction replays, and zero imbalance events. It recorded one exact failure:
`asset:world-layer` occupied unreachable terrain cell `3:2` in all 32 sessions;
13 terrain cells were reachable.

The revision policy accepts only an exact deterministic replay of that
`unreachable-asset` failure. It moved the failed asset from `(96 m, 64 m)` to
`(64 m, 96 m)`, producing region digest
`c04723a73696bb98601032e1e6d45d2e7b12bfe0db7b8ba585ae08e91d1da42a`. The after
report `3e1735863fe48157b69f647879b27c12e2c3ccac59ed12d0d608c5807757fd24` then
measured 128/128 navigation successes, 64/64 quest completions, and no failures.

The change ran through common run `run_b161fd0cb90ef0a2e9ac851d99797eee241f0b92`
using the `v7-flagship-revision` adapter. Its system actor and `full-auto`
autonomy are explicit: no human approval, playtest, review, or signature is
inferred. The four succeeded steps bind failure replay, policy admission, asset
relocation, and post-change verification. Revision record
`e3129aa4363738a27d1adfbcab592137b136d2978268cd0cbe2e59fb4c6a4544` is
independently validated against its nested region digest, deterministic
simulation replay, common-run identity, source audit artifact, before/after
artifacts, and applied change. Resealing a tampered outer record does not bypass
these checks.

## Verification

| Check                                                                                             |                      Result |
| ------------------------------------------------------------------------------------------------- | --------------------------: |
| Full `libs/isis/three-d-pipelines/src` Vitest suite                                               |                   45 passed |
| Flagship region and simulation specifications                                                     |                   10 passed |
| Common adapter and GenAI trace specifications                                                     |                    8 passed |
| Production and changed-spec strict ESLint                                                         |                      passed |
| `pnpm exec tsc -p libs/contracts/tsconfig.lib.json --noEmit`                                      |                      passed |
| `pnpm exec tsc -p libs/isis/three-d-pipelines/tsconfig.lib.json --noEmit`                         |                      passed |
| `pnpm exec nx run @oshun/contracts:build --parallel=1`                                            |                      passed |
| `pnpm exec nx run @isis/three-d-pipelines:build --parallel=1`                                     |                      passed |
| Complete generation, before simulation, common-run revision, validation, and after replay command | passed, zero after-failures |

## Published artifact identities

| Artifact                                                         | SHA-256                                                            |
| ---------------------------------------------------------------- | ------------------------------------------------------------------ |
| `libs/isis/three-d-pipelines/src/v7-flagship-region.ts`          | `abfb28d31d2d574ccc6ed6b286584d727513e530578b984572fb5e5e52dae9d9` |
| `libs/isis/three-d-pipelines/src/v7-flagship-region.spec.ts`     | `14a9af7957761d9aacc6e403f53f1a076f2ab8740b74d387e1788f7d193d3be2` |
| `libs/isis/three-d-pipelines/src/v7-flagship-simulation.ts`      | `d28380a287d0141d06afe6017b27c512cb4670004a23d1d7899c7a892dee48dc` |
| `libs/isis/three-d-pipelines/src/v7-flagship-simulation.spec.ts` | `0e1103c9f45ce3e70b19200fa06b2c6aa45b516ae7f865a153fd5120d90be87c` |

This slice changes no user-facing UI, so browser/accessibility UI automation is
not applicable. Accessibility is evaluated on the region content contract; UE
surface behavior remains unproved until a cook and playable build exist. The
simulation is not player or operations telemetry, and it does not close the
separate simulate/play/review/sign/canary bundle.
