vocal2midi-rs-review-gate

name	vocal2midi-rs-review-gate
description	Review exactly one unit and one declared role in one named Vocal2Midi Linux model manifest without writing production code.

Vocal2Midi Model Review Gate

Review one role for one unit in one selected model shard. The reviewer does not patch production code, change Cargo state, or mark a manifest promoted.

Required Context

Read the model control plane, selected manifest, unit sources/tests/assets, writer verification, dependency/bootstrap evidence, relevant historical reports, and rollback route. Do not read the closed backend manifest or other model manifests.

The review batch must contain only writer-verified reimplemented units. A review report must name the batch and record a verdict for every in-scope unit; one unit failing does not erase evidence for unrelated units.

Role Routing

• behavior_reviewer: required first for legacy_seam units; compare the named legacy public seam exactly across inputs, outputs, errors, side effects, lifecycle, and caller handoff.
• dependency_bootstrap_reviewer: verify crate-recon mode/report, features, dependency paths, asset provenance, canonical shared dependency ownership, and inventory impact.
• error_tracing_reviewer: verify structured errors, stage/path context, cancellation, redaction, and rollback diagnostics.
• data_algorithm_reviewer: verify model IO data structures, deterministic adapters, complexity evidence, and no unsupported algorithm changes.
• rust_style_reviewer: verify ownership, visibility, module shape, lints, generated-code boundary, and maintainability.
• architecture_reviewer: verify seam depth, ownership, shared capability storage, routing, and rollback.
• product_ergonomics_reviewer: verify repeated use, batching, cancellation, cleanup, recovery, and operator-facing errors.

For model_evidence execution units, do not demand output-value parity. Demand asset/load provenance, tensor names/ranks/shapes/dtypes, structured failures, repeated-call and cleanup behavior, production routing, and rollback.

Gate Order

1. Confirm the unit and review role are declared in the selected manifest.
2. Confirm writer verification passed and the unit remains reimplemented.
3. Run legacy-seam R0 first when the unit's behavior mode is legacy_seam.
4. Run the declared non-behavior role independently.
5. Record pass, pass_with_followups, fail, or not_required per unit.
6. Re-run R0 after any remediation that changes behavior, errors, schemas, lifecycle, or routing.

Do not convert a failed parity review into a Rust-only oracle. Do not treat compilation, generated code, or a Rust-to-Rust contract as promotion evidence.

Report Requirements

Write one report under rewrite-in-rust/reviews/ using the project naming convention. Include model, manifest, unit, batch, role, evidence commands, findings ordered by severity, per-unit verdict, residual risk, and rollback readiness. Reports are new evidence; existing reports remain historical and must not be edited by this gate.

Completion Response

Name the selected model/unit/batch/role, verdict, report path, highest-severity findings, checks, and coordinator readiness. Reviewers never update unit state.