Every result kwery delivers carries a confidence score (0–1) built from weighted components: identity (did we match the right product), completeness, price integrity, offer contamination, and freshness. Alongside the score, a flagged_offers array marks the specific offers that looked off. The score is absent on failed lookups, and scores below the publish threshold are held back or flagged so your repricing logic can decide how to treat them. Data quality details.
"not found" means the marketplace does not know the product. A product with no purchasable offer right now is a normal result with offers_count 0 and buybox null. Per-result fields in the docs.
0.7 is the neutral value for a product we see for the first time, when there is no earlier observation to compare against. On Google product lookups, identity is the match confidence itself. Trust scores in the docs.
For "is this the exact product", gate on trust.components.identity. score is a routing signal for overall data quality. Trust scores in the docs.
Yes. Use condition, is_direct, total and trust.flagged_offers. The buybox field follows each marketplace's own rule (on Walmart it is the lowest-priced offer of any condition), not what qualifies for you. Buybox in the docs.
kwery only emits a change event when a normalized field actually moved relative to the last validated value. Re-observations of unchanged prices are suppressed, which keeps downstream repricing logic from reacting to noise. How delta detection fits the pipeline.
Every result is resolved live from the marketplace, not pulled from a static or aggregated feed, even inside a large batch job, each SKU is fetched at the moment it's actually processed, not pre-fetched from a snapshot taken when the job started. Every record carries an observed_at timestamp and a served_from flag showing whether it came from a live fetch or cache, and you control exactly how fresh you need it with the max_age parameter (in seconds, default 1200/20 minutes): set it to 0 for a guaranteed live fetch every time, or allow a window so a request doesn't trigger a redundant fetch when nothing's likely to have changed. Learn more about our data quality.
Each record passes through identifier resolution (matching the correct product/SKU), completeness checks (no missing offers or dropped fields), and sanity checks (outlier and anomaly detection on price, currency, and availability). Anything questionable is flagged rather than silently delivered as fact.
Failures surface as explicit health signals rather than silent gaps. Records that fail validation are quarantined, the affected listing keeps its last validated value, and the full event chain is available for audit. Reliability and traceability.
Yes. Every result can carry a provenance trail showing when each step happened (fetch, match, validate, deliver), the source URL, and the outcome of each check. If a number ever looks surprising, you can trace exactly where and when it came from.
kwery normalizes seller and offer data into one consistent schema across marketplaces. Seller display variants are collapsed into canonical identities, first-party and third-party offers are tagged explicitly, and price, shipping, currency, condition, and availability use the same fields on every source, so you can compare like with like without source-specific logic.
No data provider can honestly claim 100% accuracy from public marketplace surfaces; they change constantly. What we guarantee is transparency: every record carries a confidence score, every result is timestamped and traceable, and questionable values are flagged rather than passed off as fact.
Request API access and the team will walk through sources, volumes, and integration in detail.