Interface Extraction

Reproducible provenance. Replaces any model/prompt/confidence block.

interface Extraction {
    abstained_fields: string[];
    extractor_version: string;
    input_sha256: string;
    model_artifacts: ModelArtifact[];
    parse_complete: boolean;
    reproducible: boolean;
    review_status: "ok" | "needs_review" | "human_verified";
    tier: "structured" | "text_layer" | "ocr";
}

Properties

abstained_fields: string[]

Field paths the extractor refused to guess, e.g. qa[3].answer.

extractor_version: string

Git sha (or package version) of the adapter + parser that produced this.

input_sha256: string

sha256 of the exact bytes parsed.

model_artifacts: ModelArtifact[]

parse_complete: boolean

Did every expected field extract?

reproducible: boolean

review_status: "ok" | "needs_review" | "human_verified"

tier: "structured" | "text_layer" | "ocr"