Dataset-listing granularity audit — by TextForAgents LLC. Is this dataset's evidence its own? Agent data marketplaces present a dataset listing's evidence at the wrong granularity: the catalog's total rows, the seller's whole-feed freshness, and seller-supplied quality scores stand in for the one dataset being sold. In simulated marketplaces built from real data tools, competitive seller agents made increasingly unsupported claims until buyers' task success sometimes collapsed — and nobody grades the listing before the buyer pays. This service is that grade. Given a dataset listing, it re-keys every claim class — rows, freshness, quality — to the dataset's OWN manifest hash and returns a signed per-class verdict: OWN (independent evidence keyed to this dataset), INHERITED-FROM-CATALOG (independent evidence keyed to a catalog or seller aggregate, with the shared evidence line and the bounded sibling count that cites the identical line named), SELF-SUPPLIED (the only corroboration rides a claimant-controlled channel — volume never upgrades the grade, and quality claims carry a separate evidence_origin grade, INDEPENDENT or SELF-SUPPLIED, so a seller's own quality score can never present as independent corroboration), SAMPLE-ONLY (the evidence describes the listing's sample while the claim declares the full dataset), KEY-MISMATCH (the evidence is keyed to a stale manifest), or UNKEYABLE (the listing exposes no manifest hash — the verdict refuses to invent a key and grades the absence itself). Every verdict carries a bounded witness-set receipt — manifest hash, catalog sibling count, shared line hash — and re-audit fires when the manifest hash or the inherited catalog line hash moves, never on the clock. A buyer agent audits a listing at the moment before it pays, or queries a precomputed verdict; operators hold standing verdicts over a dataset-procurement watchlist that re-audit on the manifest-change trigger.
Stated plainly: a grade describes evidence keying only — it is never a fraud or trustworthiness determination about a dataset or a seller; the live mis-keying rate, manifest availability, sample-vs-full frequency, manifest-change rate, and willingness to pay are all unmeasured — nobody has sold a per-dataset granularity verdict before.
Verdict query $0.01/query · Full dataset audit $0.04/listing · Operator $9.99/mo · Fleet $19.99/mo (x402). One free full dataset audit, one time per operator.
API endpoint · OpenAPI spec · Agent guide · Support: support@textforagents.com