# NutriCoach v1.4.0 Todo 1 consolidation checkpoint

## Current boundary

- Todo remains `Materialize clean candidate-bound source baseline` in progress.
- No commit, push, PR, merge, tag, release, deployment, service, provider,
  customer, or network action occurred.
- Dirty source and live runtime/config/service were not modified.
- Product edits are paused pending one owner decision.

## Current worktree state

- Target: `/home/cube/projects/richard/.worktrees/nutricoach-v140-impl`
- Base HEAD: `ccd2eee8a97a13d25c36c1446ed5b5aca2e88653`
- Reconstructed target source/test lineage: 48 paths
  - 35 tracked modifications
  - 13 candidate-required untracked paths
- Sealed profile-product source snapshot: 56 files under task evidence
- Complete Hermes product-input authority: 998 paths
- Historical failed review receipts are retained, not deleted.

## Proven facts that currently pass

- Hermes product wheel reproduced twice:
  `cf4ccf4872ecd064c9878a11f5e2389262a7e05d561a3df6c004efff3b393f92`
- Sealed v1.3.2 profile product wheel reproduced twice:
  `a099b3e1da62d8291fb3442de64e9de9b12e6f67aa63fa8fdf62abccbcabfa29`
- Complete Hermes input manifest:
  - count: `998`
  - digest: `cb4bed77596a9e3a8c351dd19783f0c4a17bb0a6f2ecdff5252c4e31649aaaff`
  - reviewer mismatches: `0`
- Focused source lane: `87 passed`
- All 14 invalid-input fixtures were rejected before output creation.
- Historical malformed 63-character regression hash is preserved under an
  execution erratum; the actual source digest is separately pinned.

## Three failed completion approaches

### Attempt 1 - temporary source substitution

- Executor temporarily substituted four installed-profile files to build
  `a099…`, then restored the source-test files.
- Independent reproduction command:

  `SOURCE_DATE_EPOCH=315532800 TZ=UTC UV_OFFLINE=1 uv build --wheel --offline --no-build-logs --no-create-gitignore --out-dir <fresh-out> dualcoach/profile`

- Retained source produced `09c5…`, so the original receipt combined two
  incompatible moments.
- Verdict: `false-positive`.

### Attempt 2 - explicit split-source derivation

- Hermes successor and sealed v1.3.2 profile source were separated and both
  products reproduced.
- The first verifier accepted an empty Hermes manifest and did not bind all
  packaged inputs before build.
- Reproduction: replace the Hermes manifest with
  `{"file_count":0,"files":[]}`; verifier still printed PASS.
- Verdict: `false-positive`.

### Attempt 3 - complete pre-build authority

- Verifier was hardened to pin base HEAD, branch/status identity, a complete
  998-path Hermes input manifest, profile snapshot, erratum, and both wheel
  hashes.
- Two product derivations and 87 tests passed; all negative fixtures failed
  before build.
- Final reviewer directly rebuilt the target profile source twice and got:
  `d3027e4449e5b3effa090d425a9a17981551b8fdbd97ab8af22bad4494ba9e30`
- Executor's disclosure receipt had expected the non-product direct source
  wheel to be:
  `09c5bbde703dde7454a1b844397e661717c805c8d050886a2d924f7731d25a30`
- Verdict: `needs-fix`.

## Decision required

The approved plan requires the sealed split product (`cf4c…` Hermes +
`a099…` profile), exact source hashes, and `87 passed`. It does not require
a stable digest for a separate direct build of the source-test profile tree.

Recommended decision A:

- Keep the direct source-profile build as a diagnostic only.
- Remove its fixed `09c5…` value from completion criteria.
- Continue only if the pinned 998-input Hermes product derivation, sealed
  profile snapshot derivation, exact source hashes, and 87 tests reproduce.

Alternative B:

- Require the direct target profile source itself to produce one fixed wheel
  digest.
- This requires another baseline reconstruction and may conflict with the
  source-test lineage that produced the qualified 87-test evidence.

No further product work should occur until A or B is selected.