The preloaded record is explicitly synthetic. Processing stays in this browser.
Run the example.
Boundary: a passing row does not prove factual accuracy, publisher independence, model influence, stable rank, traffic, or commercial impact. URL availability is not fetched here.
What the checker validates
Stable observation and citation IDs, a versioned prompt, answer-system context, an ISO 8601 timestamp with timezone, citation presence, an absolute HTTP(S) URL, a positive citation position, controlled source-role and publisher-relationship values, review state, taxonomy version, and explicit synthetic-data labeling.
Where this fits
Use this small implementation when validating an API object or a single annotation row. For batch research, start with the AI Answer Evidence Observation Ledger, apply the source-role taxonomy, then run the full CSV ledger validator. Citation URLs can be compared conservatively with the citation URL normalizer.
Limitations
- The checker evaluates record shape and controlled values, not whether an answer or cited claim is true.
- It does not fetch, redirect, canonicalize, archive, or score the observed URL.
- It cannot establish whether a reviewer actually observed the citation or applied the taxonomy correctly.
- AI-answer evidence remains a dated snapshot bounded by product surface, model, market, account state, and personalization.