Skip to content
KortexyaKortexyaReasoning Layer SDK

Type Alias: VerdictKind

@kortexya/reasoninglayer

@kortexya/reasoninglayer / VerdictKind

Type Alias: VerdictKind

Section titled “Type Alias: VerdictKind”

VerdictKind = { ProvenSafe: { artifactHash: number[]; binding: CertificateBinding; method: ProofMethod; proofPayload: number[]; }; } | { Falsified: { description: string; witness: number[]; }; } | { EmpiricallySafe: { confidence: number; method: EmpiricalMethod; sampleCount: number; }; } | { Residuated: { partial: number; wakeTriggers: object[]; }; }

Defined in: src/types/verification.ts:92

A verification verdict, externally tagged by its kind.

Remarks

Section titled “Remarks”
  • ProvenSafe — a certificate the engine issued. On the wire it is an UNTRUSTED transport record: the SDK does not re-check it, and an importer must run the appropriate checker on the original inputs. binding scopes the claim (kind, checker, statement and context hashes); artifactHash pins it to one artifact (model, trace, prompt schema); proofPayload is the opaque proof (SMT-LIB text, DRAT clauses, …) for the re-checker of method. Hashes and payload are byte arrays.
  • Falsified — a concrete counterexample. witness holds its canonical bytes; description is diagnostic only. Always sound.
  • EmpiricallySafe — a fuzzy or sampling score. Never sound.
  • Residuated — not decidable yet. partial is the confidence gathered so far; each wake trigger names a fact pattern whose insertion reawakens verification.

Example

Section titled “Example”
const v = cert.verdictDetail;
if ('Falsified' in v) console.log(v.Falsified.description);
else if ('EmpiricallySafe' in v) console.log(v.EmpiricallySafe.confidence);