Skip to content
KortexyaKortexyaReasoning Layer SDK

Interface: RewardScoreRequest

@kortexya/reasoninglayer

@kortexya/reasoninglayer / RewardScoreRequest

Interface: RewardScoreRequest

Section titled “Interface: RewardScoreRequest”

Defined in: src/types/reward.ts:14

Body of POST /api/v1/reward/score — score a candidate query.

Properties

Section titled “Properties”

constraints?

Section titled “constraints?”

optional constraints?: [string, string][]

Defined in: src/types/reward.ts:19

Feature → value constraints, as an array of [feature, value] pairs (e.g. [["maker","Tesla"]]). Optional.

maxInstances?

Section titled “maxInstances?”

optional maxInstances?: number | null

Defined in: src/types/reward.ts:24

Bounded sample size used the first time this tenant’s model is mined. Optional.

negativeIds?

Section titled “negativeIds?”

optional negativeIds?: string[]

Defined in: src/types/reward.ts:29

Optional explicit negatives for precision-by-example; mined from the positives when empty.

objective?

Section titled “objective?”

optional objective?: RewardObjective | null

Defined in: src/types/reward.ts:31

"recall" (default — the non-emptiness certificate) or "precision".

positiveIds?

Section titled “positiveIds?”

optional positiveIds?: string[]

Defined in: src/types/reward.ts:37

Precision by EXAMPLE: explicit positive instance ids. When non-empty (with objective: "precision") the concept is these instances, overriding targetSort.

sort

Section titled “sort”

sort: string

Defined in: src/types/reward.ts:39

The query’s sort, by lattice name (e.g. "Article", "vehicle").

targetSort?

Section titled “targetSort?”

optional targetSort?: string | null

Defined in: src/types/reward.ts:45

For the precision objective: the concept to discriminate, by lattice name. Its instances are the positives; near-miss siblings are mined as negatives.