P4: Each Harness Has Failure Modes¶

Each harness has its own failure modes.

Mitigation is asymmetric: a verifier, router, or context engine can introduce new failures of its own — for example, a verifier that hallucinates a pass. Known harness-specific failures are declared on the contract (known_failure_modes) and surfaced on the generated harness page.

Why it matters¶

Declaring a harness's own failure modes is what lets the Swiss-cheese reliability stack catch them at the next layer. If each layer pretends to be infallible, the holes line up; if each layer declares its holes, the next harness can be positioned to cover them.

How it is enforced¶

C1 forces a contract (which carries known_failure_modes), and C5: Verify on harness lets a harness detect its own failure during verification. See Failure Modes for the catalog.

Primary gates: C1: Block contract mandatory (declares the failure modes) and C5: Verify on harness (can detect them).