Repository navigation

Measure production memory delivery with auditable paired comparisons - #4

Merged

RodCor merged 16 commits into

mainfrom

codex/brain-benchmark-hardening

Sep 8, 2026

Commits on Sep 5, 2026

Make BrainBench abstention and paired comparisons honest
RodCor
committed
07265d9
Make paired benchmark failures auditable
RodCor
committed
a11285d
Finalize benchmark validation failures
RodCor
committed
d97c9a7
Measure BrainBenchmark on persistent production MCP with paired query costs
RodCor
committed
d657670
Align workflow stale scoring and terminate timed-out benchmark descendants
RodCor
committed
1c476cf
Record MCP process working set and peak memory in paired measurements
RodCor
committed
6ee3eff
Support paired reranker overrides and fingerprint the comparison runner
RodCor
committed
d39a9e7
fix benchmark model overrides surviving retrieval presets
RodCor
committed
4ac1423
isolate scored MCP retrieval from warm-start context
RodCor
committed
78f17b3
require measured session isolation in paired reports
RodCor
committed
d795cad
report stale-query denominator alongside injection rate
RodCor
committed
b3e0eda