# Phase 163 locally actionable completion evidence

Date: 2026-07-15

## Completion audit

Phase 163 defines 150 implementation requirements across eleven product areas
and ten validation requirements. The implementation is present in thirteen
focused Rust workspaces under `libs/neith/vp-*`, including the shared `vp-core`
types and `vp-testing` release-gate harness. A source audit found no `todo!`,
`unimplemented!`, ignored test, placeholder marker, or unused-code allowance in
phase-owned Rust sources.

| Phase area                  | Authoritative package | Passing tests |
| --------------------------- | --------------------- | ------------: |
| shared contracts            | `vp-core`             |            34 |
| 163.1 render cluster        | `vp-cluster`          |           109 |
| 163.2 LED wall              | `vp-wall`             |            52 |
| 163.3 camera tracking       | `vp-camera`           |            93 |
| 163.4 frustum rendering     | `vp-frustum`          |            40 |
| 163.5 colour pipeline       | `vp-color`            |            81 |
| 163.6 scene pipeline        | `vp-scene`            |            97 |
| 163.7 mixed reality         | `vp-mr`               |            97 |
| 163.8 operator control      | `vp-control`          |            75 |
| 163.9 broadcast studio      | `vp-broadcast`        |            37 |
| 163.10 capture and playback | `vp-record`           |            49 |
| 163.11 VFX handoff          | `vp-pipeline`         |            34 |
| 163.12 validation           | `vp-testing`          |            44 |

All 842 tests pass with zero failures and zero ignored cases. Every package
passes rustfmt, strict all-target clippy, and a release build. The audit applied
the previously missing canonical rustfmt output across the phase-owned sources.
No browser, mobile, or editor UI changed, so Playwright and mobile automation
are not applicable to this task.

## Open external acceptance gates

The four unchecked ledger entries are deliberately not represented as passes.
The audit hardened all locally actionable acquisition and grading work while
keeping production certification fail-closed:

| Requirement | Locally complete evidence contract                                                                                                                                                                                                                                                                                                                                               | External evidence still required                                                                              |
| ----------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------- |
| 163.12.4    | `grade_node_trace` binds the result to build/run/host, node and GPU identities, exact resolution, positive finite frame timings, configurable minimum duration, frame deadline, achieved FPS, and drop budget. Synthetic traces may meet the metric but cannot set `certified`.                                                                                                  | A live 3840×2160, 120 Hz capture of the deployed multi-GPU render node for the configured sustained duration. |
| 163.12.8    | `validate_capture` grades independently measured focus-distance, focal-length, and iris sweeps with camera/lens model and serial identities, monotonic raw counts, per-axis tolerances, minimum sampling, and live provenance. `validate_matrix` requires exactly one passing ARRI, Sony, and RED capture. It does not infer compatibility from a generic linear vendor profile. | Physical ARRI LDS, Sony, and RED camera/lens sweeps. Each vendor must produce a certified report.             |
| 163.12.9    | `grade_dogfood` verifies the Phase 27 production identity, SHA-256-addressed staged artifacts, ordered setups, unique take IDs, per-setup take counts, and live provenance.                                                                                                                                                                                                      | The production Obsidian scene/media bundle and an executed volume shoot.                                      |
| 163.12.10   | `grade_broadcast_soak` requires an identified fleet/programme, strictly increasing finite samples, configurable sample density and maximum gap, eight-hour duration, bounded least-squares memory slope, zero errors, and the drop budget.                                                                                                                                       | A real eight-hour broadcast programme run on the production fleet.                                            |

All four contracts use shared `CaptureProvenance`. A deterministic fixture is
explicitly tagged `synthetic`, can exercise pass/fail math in unit tests, and
can never produce `certified: true`. This prevents a short benchmark, lower
resolution trace, generic lens model, fabricated Obsidian completion, or sparse
two-point soak trace from being promoted to production evidence.

## On-set entry point

The release build emits `vp-validate`, a file-oriented CLI that writes a JSON
report and uses exit status 0 only for certified live evidence, 2 for valid but
non-certifying/failing evidence, and 1 for malformed inputs:

```bash
vp-validate throughput <trace.json> [policy.json]
vp-validate lens <vendor-matrix.json>
vp-validate dogfood <manifest.json> <run.json>
vp-validate soak <trace.json> [policy.json]
```

The CLI deserializes the same public, serde-backed contracts exercised by the
unit tests; there is no separate permissive parsing path.

## Verification

The following commands were run sequentially for every phase manifest from
`vp-core` through `vp-testing`:

```bash
cargo fmt --manifest-path <manifest> -- --check
cargo clippy --manifest-path <manifest> --all-targets -- -D warnings
cargo test --manifest-path <manifest> --all-targets
cargo build --release --manifest-path <manifest>
```

All commands passed. Verification was serialized and supervised on a 15 GiB host
with 31 GiB swap while an unrelated BFF development watcher was active;
available memory remained above 11 GiB. No full-workspace Nx gate was run.
