Claude dcfcb11bf7 bench: raise CI runner calibration to the full 20-session standard ...

The 8-session probe passed cleanly (0 false positives, 8/8 detection,
A/A CIs ~±0.6%); run the same 20-session protocol used for the local
trust anchor so GH-runner evidence is directly comparable.

2026-07-09 22:29:30 +00:00

..

ci.yml

bench: raise CI runner calibration to the full 20-session standard

2026-07-09 22:29:30 +00:00