Teardown: a checkout redesign that shipped twice
A composite of a failure mode we keep seeing: the spec and the research were both right.
This is a composite, assembled from patterns rather than from one company's archive. Every element of it is ordinary, which is the point — nothing here required anyone to be careless.
The setup
A team rebuilds checkout. Research is thorough. Field users make it clear that connectivity is unreliable and the cart must survive losing signal. The finding is recorded in an interview and discussed once.
The divergence
Two revisions later, the specification lists offline support as out of scope. There is no note explaining the reversal, because there was no reversal — the requirement was never carried forward. Both documents are internally consistent. Neither is wrong on its own terms.
The expensive part
It ships. Field usage collapses in exactly the way the interview predicted. The team rebuilds the flow, and the second version implements the requirement that was established before the first version started.
What would have caught it
Not more review. Three careful people read both documents and neither contradicted itself. What was missing was anything comparing the specification against the research it was supposed to be based on — a question nobody owns, because it falls between two artefacts rather than inside one.
Stop writing PRDs without understanding the product.
Anchor reads your interviews, tickets and docs first — then writes, citing every line.
