Price and delivery describe the offer. Trust describes whether it will go wrong. For an agent acting with real money, the second matters more.
Review scores mostly measure how recently a seller bought reviews. Observed behaviour over time — fulfilment record, delisting rate, stock accuracy, how long they have existed — is harder to fake and more predictive.
`price_implausible` fires when an offer sits far below the product's own distribution across every seller. For an agent optimising a budget this is the single most valuable signal in the API, because that offer is usually a different item, a different region, or a listing that will be cancelled after you order.
{ "risk": "elevated",
"flags": [
{ "code": "price_implausible", "severity": "high",
"detail": "68% below the 90-day 5th percentile across all sellers" } ],
"recommendation": "Do not auto-purchase. Surface to the user with the flags." }Where a brand publishes no authorised-seller list, authorisation returns `unknown`. Reporting absence of evidence as evidence of absence would make the field useless in exactly the categories where it matters.
A 15% restocking fee on a $200 item is $30 of risk — frequently more than the price difference an agent was optimising over. Read `/v1/offers/{id}/returns` before treating two offers as comparable.