Answer Card Projection#

buildAnswerCardProjection(report, claimId) is the compact, report-only read model for a product answer card. It accepts an already-derived TrustReport; it does not validate untyped input, rerun status derivation, calculate freshness, or infer support.

import { buildAnswerCardProjection } from "@kontourai/surface";

const card = buildAnswerCardProjection(report, claimId);

The result is discriminated by found.

• A found card preserves the exact claim subject, unknown claim value, derived status, copied freshness facts, materiality, claim-scoped transparency gaps, and evidence associated with that claim.
• A missing card is stable: claim is null, both evidence buckets and direct inputs are empty, derivation is unavailable, and gaps are empty.

Evidence is partitioned with partitionEvidenceBySupport: entailing evidence first retains its original relative order, then cited evidence retains its original relative order. Legacy evidence that omits supportStrength belongs in the entailing bucket but exposes supportStrength: null, making that missing producer fact visible. result reads only Evidence.passing: true is "passed", false is "failed", and an omitted value is "not-evaluated". A failed entailing record blocks only when blocking !== false; cited evidence never blocks the claim regardless of its blocking field.

Derivation is one level deep. Direct inputs use derivationInputsForClaim, so edge declarations precede legacy derivedFrom declarations and duplicate input IDs keep the helper's declared-order behavior. Missing input claims remain in the card with status: null. If an unexpected direct-input projection fault occurs, the card reports derivation.available: false and empty direct inputs without discarding the claim, evidence, or transparency gaps.

The projection has no JSON Schema. It is a TypeScript read model over the existing TrustReport, just like other report projections; portable format schemas remain the Hachure-derived files under schemas/.