docs/testing/2026-09-30-instruction-trim/clankie-suite-codex-summary.md

clankie · codex codex-cli 0.159.1 · gpt-6-astra

ArmPass rate95% CITokens/trial (95% CI)Wall/trial (95% CI)TrialsErrors
pre-145648/48 (100%)93% to 100%64k (57k to 70k)26 s (22 s to 30 s)480
trimmed48/48 (100%)93% to 100%47k (42k to 52k)25 s (20 s to 29 s)480

trimmed − pre-1456, paired by case (case-clustered bootstrap):

MeasureDifference95% CIVerdict
Pass rate+0%+0% to +0%within noise (16 cases)
Pass rate: heldout+0%+0% to +0%within noise (5 cases)
Pass rate: incident+0%+0% to +0%within noise (5 cases)
Pass rate: social+0%+0% to +0%within noise (6 cases)
Tokens/trial−17k−27k to −11koutside noise (16 cases)
Wall/trial−1 s−3 s to +1 swithin noise (16 cases)