Compare a candidate PDF against the original it claims to be

Accepts multipart/form-data with two file fields: original (the reference: the file you generated or hold on record) and candidate (the file being verified). Returns a CompareReport: byte identity, revision ancestry and structural difference, as facts.

The three tiers, strongest first.

1. relationship: "identical": the files hash equal. The only result that is a proof.
2. "candidate-extends-original" / "candidate-truncates-original": one file contains the other verbatim up to a revision boundary. PDF edits-in-place append rather than rewrite, so the appended revisions ARE the edit; the report classifies the objects they touch (ancestry.contentTouched is the headline fact: page/content/image objects were replaced after the shorter file was complete).
3. "rewritten": neither contains the other. The report degrades honestly to parsed structure: Info/XMP metadata field by field, the /ID pair (permanent half vs per-save half), fonts added/removed, page count and page geometry.

What it cannot do. No pages are rendered, so no result asserts the rendered appearance is identical. A "rewritten" pair with clean structure is still only structurally clean. Findings are facts with severities, never verdicts.

Cost. One comparison parses two documents and is metered as 2 documents against the monthly quota. Requires an API key (or a dashboard session); there is no anonymous access to this endpoint.