Answered without a chart tool call, so the field was re-asked; distinct from an honest no_retrieval_results blank.
Reported answered but gave nothing the field's type could carry.
A filled field with sub-fields that got no value. Counts only: naming them would disclose by complement what the chart said.
The list answer had more items than the field has boxes; the surplus was dropped.
A filled text answer longer than its box holds; drawn in full, so the reviewer must see it.
The first answer ran past the box; the re-ask's shorter answer replaced it.
A non-fatal problem worth surfacing on the run. v1 swallowed every one of these, so a run whose retrieval returned nothing looked exactly like a clean one in the database.