Rulvar API reference / @rulvar/evals / RunSweepOptions
Interface: RunSweepOptions
Defined in: packages/evals/src/sweeps.ts:56
Properties
| Property | Type | Description | Defined in |
|---|---|---|---|
committerId | string | The dedicated committer identity. | packages/evals/src/sweeps.ts:60 |
engineFor | (member) => | Engine | Promise<Engine> | A fresh engine per model cell, routed at that member: the caller owns adapters, budgets, and the VCR posture, so a sweep records and replays like any engine run. | packages/evals/src/sweeps.ts:68 |
envelope? | SpendEnvelope | Aggregate debit-only envelope over the WHOLE matrix (v1.16.2 review P1-2): every target and judge run authorizes its immutable ceiling before starting, so the pool times cases times judge-call product cannot exceed it, falsification pool growth included. An envelope requires suite.budgetUsd (and suite.judgeBudgetUsd once a grader judges). Refusals are monotone (v1.17.0 review P1-5): a refused target stops that cell's walk but everything already measured stays on the cell (n, costs, caseNames), judge refusals normalize into their row's incomplete marker, and an incomplete cell emits NO claim. Share the instance with the canary loop so probes draw from the same remainder. | packages/evals/src/sweeps.ts:86 |
modelEpochFor? | (member) => | { canaryFingerprint?: string; capsHash?: string; pricingVersion?: string; registryVersion?: string; } | undefined | Optional epoch stamp per pool member (capture via the core modelEpochOf; the canary fingerprint rides it when probes ran). | packages/evals/src/sweeps.ts:93 |
observedAt | string | ISO date of the sweep; the TTL table applies from it (no wall clock inside). | packages/evals/src/sweeps.ts:62 |
reportId | string | Deterministic, caller-minted; every claim's evidence and gate reference it. | packages/evals/src/sweeps.ts:58 |
store? | ModelKnowledgeStore | When given, emitted claims commit through the committer identity. | packages/evals/src/sweeps.ts:88 |
suite? | RunEvalSuiteOptions | Passed through to every suite run (budget, VCR hooks ride the engine). | packages/evals/src/sweeps.ts:72 |
thresholds? | Partial<SweepThresholds> | Mid-band pass rates emit NO claim (uninformative); see defaults. | packages/evals/src/sweeps.ts:70 |