Expand description
“Up to N examples, and the count of what they stand for” — one collector (RFC 09 §5.1 O6).
Every judge in this crate names a handful of offenders and then says how
many more there were: the doctor’s per-check findings, field’s per-path
ones, expect’s violations, cutover’s leaked keys, budget’s example
expansions, why’s evidence lines. It was written out by hand at each
site, with a different cap and a slightly different closure, and the deep
review found what that costs: the qos-observed-mismatch cap capped the
candidates rather than the findings, so violators past the first twenty
judged keys vanished and the remainder note under-counted them.
The bug is structural — a hand-written cap counts whatever it happens to
be looking at — so the fix is one collector that counts everything offered
and keeps the first cap. The cap value stays each site’s own policy;
only the mechanism is shared.