Phase 3 — structured L3 review panels #9

Scope (ROADMAP.md, Phase 3)

• Role-specialized reviewer schemas: statistician, domain theorist, methodologist, devil's advocate.
• Multi-round reports, objection tracking, disagreement gates.
• Catch-rate scoring on seeded known-bad and known-good candidates.

Acceptance checklist

• Structured reviewer schemas for the four roles, each emitting typed verdicts (approve / object / abstain) with severity-graded objections and rationales.
• Multi-round panel protocol: independent round 1, rebuttal rounds on disagreement, objection lifecycle tracking (raised / withdrawn / sustained).
• Disagreement gates: blocking objections prevent L3 promotion until resolved; deterministic reference reviewers for the synthetic domains.
• Panel verdict is recorded as review evidence through the kernel/ledger (no side channels).
• Catch-rate scoring harness over Phase-2 seeded fixtures: report known-bad catch rate and known-good false-reject rate.
• Tests green (ruff + pytest 3.10/3.11/3.12). Docs updated truthfully.
• Invariants: L4 human-gated; no discovery claims; no secrets; $0 spend.

Dependencies

Blocked by #8 — Phase 2 epic (panels consume validation gates, seeded fixtures, and calibration reports).

Tasks

• P3-T1: reviewer role schemas #23 reviewer role schemas
• P3-T2: multi-round panel protocol + objection tracking + disagreement gates #24 multi-round panel protocol + objection tracking
• P3-T3: deterministic reference reviewers (synthetic domains) #25 deterministic reference reviewers
• P3-T4: catch-rate scoring harness #26 catch-rate scoring harness
• P3-T5: kernel L3 panel gate (opt-in) #27 kernel L3 panel gate (opt-in)
• P3-T6: Phase 3 docs + version 0.4.0 #28 Phase 3 docs + version 0.4.0