Evidence-Based Verification

Multi-tier verification checks and cryptographic completion evidence.

In M31A, no task or mission is marked complete merely because the model claimed it succeeded. Completion strictly requires multi-tier cryptographic verification evidence.

Evidence-Based Completion

The LLM cannot declare success. The CompletionGate evaluates structured outputs from compiler checks, unit test suites, linters, and semantic reviews before certifying mission state.

Multi-Tier Quality Gates

Tier 1: Syntax & Compilation

cargo check / rustc validation

MANDATORY

Tier 2: Automated Unit & Integration Tests

cargo test execution with zero failures

MANDATORY

Tier 3: Static Analysis & Linters

cargo clippy -- -D warnings, cargo fmt --check

MANDATORY

Tier 4: Anti-Fake-Diff Review

Rejects todo!(), unimplemented!(), and stub code

MANDATORY

Cryptographic SHA-256 Digest

Every verification result produces a content-addressed SHA-256 evidence digest that is permanently recorded in the SQLite checkpoints table.