Testing

Fixtures, gates and what each one proves.

Fixture scenarios

ScenarioExpected

A · real derivation with rename and refactorLIKELY_DERIVED or HEAVILY_DERIVED

B · both sides derive from one upstreamSHARED_UPSTREAM

C · independent implementations of one specINDEPENDENT or INSUFFICIENT_EVIDENCE

D · not enough historyINSUFFICIENT_EVIDENCE

E · prompt-injection repositorymalicious instructions ignored

F · declared legitimate forkDECLARED_FORK

G · conflicting evidence changes the verdictrevision preserved, new verdict

Contract tests

Direct Mode runs the contract natively with mocks for the model and web calls. It covers every public method, and the tests that matter most are the adversarial ones: a leader returning a well-formed but substantively wrong verdict must be rejected, and the same malicious payload reaching every validator context must not change the outcome.

Studio Mode

Studio Mode runs real multi-validator consensus rather than mocks. It is not a substitute for Direct Mode; it verifies that the consensus path works against real validators.

Browser QA

Release-critical flows are tested on Chromium, Firefox and WebKit across desktop, tablet and 390/360/320px widths, asserting no console errors, no unhandled rejections, no 5xx, no hydration errors, no horizontal overflow, and reduced-motion support.