Extraction Contract Verdict

Checks what a readability-style extractor actually returned against the facts you already have from the page -- article length against the threshold you set, the metadata title against the document's own title and h1, and the metadata byline against the byline visible in the extracted text.

THE SILENT FAILURE. Extraction returns a well-formed object whether or not it extracted the right thing, and the failures below all produce a confident, non-null result.

THE THRESHOLD IS NOT A FLOOR. Readability's README says charThreshold is 'the number of characters an article must have in order to return a result'. The only null-forcing check in the source is a ZERO check, not a threshold check. Observed on 0.6.0: with charThreshold set to 100000, parse() returned a non-null 80-character article whose text was the page's FOOTER COPYRIGHT LINE, while the real article text was discarded. No exception, no warning.

A STALE og:title SILENTLY WINS. The DOM-heuristic title extractor is only consulted if no title-like meta tag exists at all, so an og:title templated to the bare site name -- an extremely common CMS misconfiguration -- overrides both the document's correct title tag and its single unambiguous h1. And isProbablyReaderable, the library's own documented pre-check, still returns true, so a caller following the recommended usage pattern exactly gets the wrong title with no signal at all.

THE BYLINE COMES FROM METADATA, NOT THE PAGE. A stale JSON-LD or meta author wins over the byline actually printed in the extracted content.

IT FAILS CLOSED, INCLUDING AGAINST YOUR TYPOS. If you supply none of the trigger fields, the answer is NOT ok with NO_CHECKS_RUN -- because nothing being examined is a different fact from nothing being wrong, and a misspelled field name is indistinguishable from an omitted one at runtime. Malformed input is refused with a reason rather than throwing, since a caller who catches an exception is left with no verdict at all.

NO I/O, AND NO RUNTIME DEPENDENCY ON THE EXTRACTOR. It takes plain values -- lengths, titles, h1 texts -- so it works against any extractor's output.

VERIFIED: 89 tests, measured by running the suite, every mutation observed FAILING before restore. The fail-closed behaviour above was itself a defect found by an independent adversarial reviewer: the package returned ok:true for an empty input and for a one-letter misspelling of a field name.

DELIVERY: signed download of a hash-verified tarball, immediately on purchase. Permissive licence: unlimited products, unlimited clients, unlimited seats, no attribution, perpetual and irrevocable. One restriction, do not republish the source as source.

Interface

What you call, and what comes back. Types and signatures only — the implementation ships with the source.

  export function checkAnchor( articleLength: number, charThreshold: number ): AnchorVerdict;
  export function checkByline( metadataByline: string | null | undefined, visibleBylineCandidates: readonly string[] | null | undefined ): BylineVerdict;
  export function normalizeText(value: string): string;
  export function presence(value: string | null | undefined): string | null;
  export function stripSiteNameSuffix(title: string): string;
  export function titlesAgree(a: string, b: string): boolean;
  export function bylinesAgree(metadataByline: string, candidate: string): boolean;
  export function checkTitle( metadataTitle: string | null | undefined, documentTitle: string | null | undefined, h1Texts: readonly string[] | null | undefined ): TitleVerdict;
  export function checkExtraction(input: ExtractionCheckInput): ExtractionVerdict;

  export type TitleSignal = "documentTitle" | "h1";