Record the revised registry benchmark result (#2261)

The first #2260 benchmark run exposed eager alias indexing during
ingest.
The revised commit passed CI Benchmarks after index construction moved
to
alias lookup. Keep the owning subsystem contract aligned with that
result.

Refs #2199
This commit is contained in:
rcourtman
2026-09-24 21:37:33 +01:00
committed by GitHub
@@ -5383,4 +5383,5 @@ alternating exact-base/candidate worker runs at load 0.53-0.91 measured the
versus 4.953ms (+0.7%), with unchanged allocations. The 200-host merge was
3.210ms versus 3.192ms (-0.6%). The warmed alias-hit benchmark remained about
4.7µs and 19 allocations. These paired runs remove the observed local ingest
regression; the revised PR still needs its CI benchmark result.
regression. The revised #2260 CI Benchmarks job passed on commit
`bbce69ea65` with no benchmark regression gate failure.