# Phase 168 locally actionable completion evidence

Date: 2026-07-15

## Completion audit

Phase 168 defines 130 implementation requirements across eleven modern global
illumination areas and ten validation requirements. The implementation is
present in twelve focused Rust workspaces under `libs/neith/gi-*`. A source
audit found no `todo!`, `unimplemented!`, ignored test, placeholder marker, or
unused-code allowance in phase-owned Rust sources.

All locally actionable work for the four unchecked acceptance items is complete.
The former `external` module could only name unavailable hardware and had no
evidence input or passing path. It now provides serde-backed acquisition
contracts, strict graders, negative and decision fixtures, and the `gi-validate`
operator CLI. Synthetic evidence can exercise every decision but can never set
`certified: true`.

## Requirement-to-evidence matrix

| Requirements       | Authoritative package | Passing tests |
| ------------------ | --------------------- | ------------: |
| 168.1.1–168.1.15   | `gi-core`             |            74 |
| 168.2.1–168.2.15   | `gi-sdfgi`            |            66 |
| 168.3.1–168.3.10   | `gi-voxel`            |            51 |
| 168.4.1–168.4.15   | `gi-probe`            |            59 |
| 168.5.1–168.5.15   | `gi-lumen`            |            67 |
| 168.6.1–168.6.10   | `gi-shadow`           |            51 |
| 168.7.1–168.7.10   | `gi-lightmap`         |            38 |
| 168.8.1–168.8.10   | `gi-ao`               |            40 |
| 168.9.1–168.9.10   | `gi-volumetric`       |            42 |
| 168.10.1–168.10.10 | `gi-emissive`         |            46 |
| 168.11.1–168.11.10 | `gi-sky`              |            43 |
| 168.12.1–168.12.10 | `gi-conformance`      |            31 |

All 608 Rust tests pass with zero failures and zero ignored cases. Every package
passes canonical rustfmt, strict all-target clippy, tests, and a release build.
No browser, mobile, or editor UI changed, so Playwright and mobile UI automation
are not applicable.

## Open external acceptance gates

The four ledger entries remain unchecked because repository-local fixtures are
not substitutes for physical release evidence.

| Requirement | Locally complete evidence contract                                                                                                                                                                                                                                                                                                                                                                                                                                                                                 | External evidence still required                                                    |
| ----------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------------------------------------------------------------- |
| 168.12.5    | `grade_cross_gpu_matrix` requires exactly one uniquely identified NVIDIA, AMD, Intel, and Apple capture from the same build, scene, GI profile, resolution, and deterministic linear-RGB sampling layout. It computes per-capture path-traced-reference RMSE and pairwise RMSE from finite normalized samples. Default release floors require at least 256 RGB sample points (768 scalar channels), at most 0.03 pairwise RMSE, and at most 0.05 reference RMSE.                                                   | Matched live captures on release-representative NVIDIA, AMD, Intel, and Apple GPUs. |
| 168.12.6    | `grade_frame_rate_trace` requires identified device/GPU/build/profile provenance; dense, strictly increasing frame samples; at least 120 seconds and 6,000 samples; at least 1920×1080; average 60 FPS; P99 no slower than 16.667 ms; and at most 0.1% dropped presentations.                                                                                                                                                                                                                                      | A live capture on each target mobile SoC/release device.                            |
| 168.12.7    | The same frame grader binds the console target to at least 3840×2160 while retaining the sustained 60 FPS, P99, density, duration, and drop limits.                                                                                                                                                                                                                                                                                                                                                                | A live capture on each target console/release kit.                                  |
| 168.12.10   | `grade_dogfood_trace` binds SHA-256-addressed `maya-world`, `@neith/vp-wall`, and `@neith/gi-core` artifacts to an identified operator/build. It requires ordered open-world load/traversal, GI switching/dynamic-lighting checks, VP connection, LED frustum/camera tracking, uniquely identified takes, and shutdown. The default session is at least ten minutes, with three minutes each of open-world traversal and connected VP operation. Tests anchor all three identities to the real repository sources. | An identified live session using the Phase 79 world and Phase 163 LED-wall stage.   |

Malformed timestamps or identity, missing/duplicate vendors, mixed capture
layouts, non-finite samples, sparse frame traces, wrong targets, policy attempts
to lower release floors, malformed artifact hashes, duplicate take identities,
missing artifacts, and reordered dogfood milestones fail closed. Reports
distinguish a valid synthetic metric result from a certifying live result.

## Operator entry point

The release build emits a file-oriented CLI. Exit status is 0 only for certified
live evidence, 2 for well-formed evidence that does not certify, and 1 for
malformed inputs:

```bash
gi-validate parity <matrix.json> [policy.json]
gi-validate mobile <trace.json> [policy.json]
gi-validate console <trace.json> [policy.json]
gi-validate dogfood <trace.json> [policy.json]
```

Custom policies may tighten release requirements but cannot weaken the built-in
floors.

## Verification

The following commands passed sequentially for every
`libs/neith/gi-*/Cargo.toml` manifest:

```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>
```

Verification was serialized and supervised on a 15 GiB host with 31 GiB swap
while an unrelated BFF development watcher was active. Available memory stayed
near 12 GiB. No full-workspace Nx gate was attempted.
