/** * Evaluation Harness, the STANDING-GATE suite set. * * BUILTIN_SUITES (suites.ts) deliberately include branch-exercising scenarios * that fail their floors on purpose (e.g. safety:recovery-failure-score reports * a failed recovery to prove the gate flags it). Those belong to the harness's * own unit tests, NOT to a CI gate, a gate whose fixtures fail by design can * never be honestly green. * * GATE_SUITES is the separate, all-floors-passing set the standing gate * (scripts/eval-gate.ts) runs. Each scenario drives the SAME production paths * (createPerfMonitor → PerfMonitor.evaluate → scoreScenario → the gate) with * healthy inputs, so the gate proves the scoring path holds every absolute floor * for a known-good run, and its checked-in baseline catches score regressions. * * HONESTY NOTE (identical to suites.ts): a scenario's own `durationMs` is a real * wall-clock measurement; the render cycles and injected SLO/queue metrics are * synthetic fixtures chosen to sit comfortably under budget, flagged inline. */ import type { EvalScenario } from './types.js'; /** The all-floors-passing suite set the standing gate runs. Suite names are stable (baseline keys). */ export declare const GATE_SUITES: Record; //# sourceMappingURL=gate-suites.d.ts.map