Type Alias: ClaimVerdict

@kortexya/reasoninglayer / Guardrail / ClaimVerdict

Type Alias: ClaimVerdict

ClaimVerdict = { certificateId: string; citations?: ClaimCitation[]; ruleProgramHash?: string | null; ruleRevision?: number | null; status: "certified"; } | { reason: string; status: "refuted"; } | { reason: string; status: "abstained"; }

Defined in: src/types/guardrail.ts:87

The disposition of a single claim.

Type Declaration

{ certificateId: string; citations?: ClaimCitation[]; ruleProgramHash?: string | null; ruleRevision?: number | null; status: "certified"; }

certificateId

certificateId: string

Content-addressed id of the persisted soundness proof.

citations?

optional citations: ClaimCitation[]

Supporting sources recovered from the proof.

ruleProgramHash?

optional ruleProgramHash: string | null

Hex SHA-256 of the rule program active when the proof was issued.

ruleRevision?

optional ruleRevision: number | null

In-process rule-program revision ordinal, when tracked.

status

status: "certified"

{ reason: string; status: "refuted"; }

reason

reason: string

Why the claim was refuted.

status

status: "refuted"

{ reason: string; status: "abstained"; }

reason

reason: string

Why the guardrail abstained.

status

status: "abstained"

Remarks

A discriminated union on status:

certified — the claim was proven against the knowledge base; the soundness proof is persisted under certificateId and its supporting sources are recovered into citations.
refuted — the claim contradicts the knowledge base (a caught hallucination).
abstained — the guardrail declined to rule on the claim.

Example

if (verdict.status === 'certified') {
  console.log(verdict.certificateId, verdict.citations);
} else {
  console.log(verdict.reason);
}