Two ways the harness lied about what it was measuring.
The router answers an index request with a redirect, and fetch() did not
pass -L, so a --live byte comparison compared against a 302 body rather
than the index a client ends up with.
A purge is asynchronous, so a run started right after a reindex read
whatever the edge still held. That produced a phantom '161 regressions'
on amd64/alpine323 whose PACKAGES.rds was already correct - the .gz was
simply a stale copy, and re-measuring after the purge settled showed
zero. A verification tool that races the thing it verifies is worse than
no tool.
pat-s
2171f8ea3b