Knockoff Data API v1

matched the brand's marks trace to one distinct owner. ambiguous several distinct owners use this name; we state the marks but claim no owner. no_record no USPTO mark for this string (itself a signal).

live dead none best status among the matched marks.

Present only when a resolve call's title/URL carried product context. The brand is ambiguous register-wide (register_state), but exactly one owner holds its marks in the queried Nice classes — e.g. {"classes": ["021"], "context": "cookware", "register_state": "ambiguous"}. The verdict holds for those classes only: do not cache or reuse it as an unconditional answer for the brand.

One operator holds a calibrated supermajority of the brand's live marks — e.g. {"share": 0.818, "live_marks": 108, "of_live_marks": 132, "other_candidates": [...]} — and is reported as the owner (when the call carried product context, only if that operator also holds live marks in the queried classes). other_candidates discloses every minority filer by name and live-mark count; weigh them before treating the owner claim as exclusive.

The seller's identity card, beside the brand's facts. domain (the merchant's identity key), platform + platform_confidence (deterministic = a URL shape the platform doesn't let merchants change, e.g. Shopify's /products/{handle}?variant=; heuristic = a default path shape other platforms can collide with), trademark_footprint (the store name's own facts object), and identity_completeness — which card fields are filled, a coverage statement, never a judgment.

domain_age (months since the domain's RDAP registration date — young domains are the fly-by-night signal), legal_entity (the operating entity named on the site's own legal pages, always with its receipt: {"name", "source_url", "quote"}), and a confirmed platform_confidence are filled by async enrichment: the first resolve of a domain queues it (enrichment: "queued"), later resolves return the enriched card ("complete", or "failed"). Enriched fields may be up to 30 days old — a domain is never re-fetched inside that window.

Present when the URL carries an Amazon seller token (/sp?seller=, or seller=/smid= on a product URL): the third-party seller behind the listing. legal_name, country, address_present, and registration_number_present come from the seller's INFORM Consumers Act disclosure as observed on-page (captured_at is the receipt; corroborated = the address was confirmed from ≥2 distinct networks). entity is the seller's shared-address cluster (how many storefronts operate from the same address).

trademark_footprint appears only when the legal name joins a trademark operator under the stated join_basis (entity_country = distinctive entity name + matching country — never a bare name match); it is the seller's own portfolio (distinct_brands, live_marks, mill_filed). An unknown seller returns the id with nulls — the miss is recorded and coverage grows with observation.

Is the seller of record the brand's operator? sells_direct the seller's disclosed identity and the brand's resolved owner fold to the same entity (distinctive entity name + matching country, the same corroboration bar as join_basis: entity_country), or their registered aliases share a published operator cluster; the basis states entity_country or published_cluster. third_party both identities are comparable and the entities differ. platform_direct the marketplace itself is the seller of record (e.g. Amazon retail's own seller id). unknown nothing provable, with the reason stated (no_seller_token, seller_unresolved, brand_owner_unresolved, identity_not_comparable, or country_mismatch — a same-named entity in a different country proves neither sameness nor difference, so it stays unknown). seller_entity / brand_entity carry the two compared identities as receipts. Never a score.

Is the listing what it claims to be? Composed from crowd-relayed observations (extension sightings, hardened by a 2-network corroboration gate) plus our own reverse-image receipt over the listing's hero image. status is observed or the honest no_observation (coverage is demand-shaped — unseen listings have no data, and none is invented). claim carries the crowd-corroborated identity tuple (attribution, ships-from, duties-paid language, AI-scale flag); hints carries uncorroborated latest-value numerics (sales, shop age); signals lists human-checkable reasons with their evidence grade. receipt is the reverse-image result (catalog_matches = mass-market catalog twins, always with caveat: "copy_direction_unverified" — copycat sites also steal from real artisans). lean is advisory and watch is its ceiling — “worth a human's look, here are the receipts” (lean_basis), never a confident “fake” label. Never a score.

ISO-2 country of the current owner. Only present when matched — an ambiguous brand never gets an owner claim.

Months since the earliest real filing among the matched marks.

Recorded sales of the mark (true ownership changes — renames and collateral liens don't count).

Name of a published operator cluster this brand belongs to. Clusters publish only when public records or calibrated matching rules clear the measured publication bar; the audit basis is returned in operator_footprint.cluster.basis.

How many marks matched, how many live, and on what basis (wordmark / pseudo_mark).

Up to 20 serial numbers with TSDR links — every claim is independently verifiable.

How many distinct brand names this operator holds across the whole register. The headline: a real brand files a handful for its own products; a mass operator files dozens. Counted over the operator's resolved identity — its name variants folded together (see confidence).

The owner's live and total mark counts.

The owner's filing window. 8 brands across a 4-month burst reads very differently from 10 brands across 12 years.

company individual null from the USPTO legal-entity code; an individual holding many brands is a strong mass-filing signal.

True if any of the owner's marks were filed through a high-volume filing attorney whose book is ≥50% CN/HK owners. Most meaningful for small-footprint owners; a century-old global filer can flag on one mark.

{known, size, basis} — whether this operator belongs to a published cluster, how many brands it spans, and its audit trail (record, ai, human, or mixed). Only returned when matched.

resolved_entity the owner's name variants (“Procter & Gamble Company” / “The Procter and Gamble Co.”) were folded into one operator, corroborated by a shared country, so the counts are the operator's true total. exact_name no variant was found, so the counts are keyed on the single filing string and are a floor: an operator filing under variants we couldn't corroborate is undercounted. The fold is a deterministic name+country match, separate from the published cluster inference.

How many distinct owner-name strings were folded into this resolved operator. 1 means none were found (so the counts equal the exact-name floor); higher means variants were merged.

On ambiguous brands the footprint shown is the most prolific candidate owner's; this is how many other candidates exist. 0 when matched.