The missing field is the product: why agent listings need declared, derived and unknown

The taxonomy names three classes and the sample record uses two. Declared values, decision flags, a list of unknowns and a checked_at date are all present; there is no derived block anywhere in the illustrative shape ^[19]. That absence is the interesting part, because derived is the class doing the difficult work: it is where a platform computes a value and stays accountable for it, on condition that the rule is visible and the inputs are named ^[4]. Declared and unknown are bookkeeping you can implement in an afternoon. Derived is the part that has to be built, and the part that gets read back to you in a dispute.

The declared numbers in the sample reward a second look. Handling time is 72 hours, delivery to Singapore is estimated at 12 days with a default of 18, and the return window is 7 days ^[6]. Add handling to delivery and a Singapore buyer is waiting about 15 days, everyone else about 21 ^[18]. A 7-day return window sitting against a 15-day wait either starts at delivery or is close to meaningless, and the record does not say which. This is the case for representing return terms as conditions instead of an "easy returns" phrase ^[15]: two declared fields can contradict a third with no model error involved at all.

Note which fields the sample marks unknown. Exact dimensions, weight, supported laptop size range, maximum load, included parts, material thickness ^[7]. Those are the fields a buyer uses to decide whether the thing fits the laptop they own. An agent that quietly supplies a supported size range has not padded a description, it has issued a compatibility claim that no accountable party made ^[3]. Hence the reason codes: not_declared_by_seller routes to ask_seller and blocks the compatibility recommendation outright ^[9], and stale_value, conflicting_sources, not_applicable and outside_sale_region each imply different handling ^[10]. A single null cannot carry any of that.

The same discipline applies to the flags. New seller and no sales history are in the sample as evidence for a human to weigh, not as a verdict, and the post is explicit that sparse history must not be converted into a manufactured reputation score ^[11]. WebAZ's surface exposes one search tool and cannot create orders or move funds ^[12], which is the cheap moment to argue about this. Nothing downstream of a wrong answer is irreversible yet.