Skip to content

Interface: RewardScoreResponse

@kortexya/reasoninglayer


@kortexya/reasoninglayer / Reward / RewardScoreResponse

Interface: RewardScoreResponse

Defined in: src/types/reward.ts:49

Response of POST /api/v1/reward/score.

Properties

certifiedNonEmpty

certifiedNonEmpty: boolean

Defined in: src/types/reward.ts:55

A sound certificate — recall: provably non-empty (subsumes ≥ 1 populated anchor); precision: provably discriminating (covers ≥ 1 positive, 0 negatives). Never a false positive.


coversNegative?

optional coversNegative: number | null

Defined in: src/types/reward.ts:57

Precision objective only: negatives (mined near-miss siblings) covered.


coversPositive?

optional coversPositive: number | null

Defined in: src/types/reward.ts:59

Precision objective only: positives covered.


estimatedExtent

estimatedExtent: number

Defined in: src/types/reward.ts:61

Recall: a sound lower bound on the query’s extent. Precision: positives covered.


precision?

optional precision: number | null

Defined in: src/types/reward.ts:63

Precision objective only: fraction of covered instances that are positives.


recall?

optional recall: number | null

Defined in: src/types/reward.ts:65

Precision objective only: fraction of the concept’s positives covered.


reward

reward: number

Defined in: src/types/reward.ts:70

Recall: ln(estimatedExtent + 1). Precision: the F1 of discriminating the concept. A bounded RL reward signal in both cases.