Skip to main content

EvalCaseResult

@forge/monorepo


@forge/monorepo / backend/src / EvalCaseResult

Type Alias: EvalCaseResult

EvalCaseResult = object

Defined in: backend/src/persistence/index.ts:809

One case's result, with everything needed to reproduce it.

The grader id and version, and for a judged case the model and prompt version, are stored on the result. A run that recorded only scores could not answer "why did this change" — a prompt edit and a model upgrade look identical in the numbers, and both look like a quality change.

Properties

caseId

readonly caseId: string

Defined in: backend/src/persistence/index.ts:810


costMinorUnits

readonly costMinorUnits: number

Defined in: backend/src/persistence/index.ts:820

Integer minor units. Zero for a deterministic grader, which is asserted rather than assumed.


dimension

readonly dimension: string

Defined in: backend/src/persistence/index.ts:811


expectKind

readonly expectKind: string

Defined in: backend/src/persistence/index.ts:812


graderId

readonly graderId: string

Defined in: backend/src/persistence/index.ts:814


graderVersion

readonly graderVersion: string

Defined in: backend/src/persistence/index.ts:815


modelId?

readonly optional modelId?: string

Defined in: backend/src/persistence/index.ts:817

Present only for a judged case. Absent means graded by code, which is also the cheap path.


promptVersion?

readonly optional promptVersion?: string

Defined in: backend/src/persistence/index.ts:818


verdict

readonly verdict: EvalVerdict

Defined in: backend/src/persistence/index.ts:813