topbot

Data · Runs

Every campaign run.

Each row is one invocation of the runner. A run of Claude Code is one task attempted by one arm. Runs that never started are not counted. Cost is the API price of the tokens with cache writes at the five-minute rate; the effort page prices them at the one-hour rate.

DateCampaignWhatOutcomeRunsTokensCost
2026-10-05CA1Trial of three (kill switch). A forced skill-evaluation hook vs no hook, skill-activation bank180.3 m$0.62
2026-10-05CA1A forced skill-evaluation hook vs no hook, skill-activation bankactivation +30.5 points [17.1, 43.9]2003.3 m$7.81
2026-10-05C0Calibration of TB-R (review bank)item windows, ICC, spend per cell15411.9 m$13.83
2026-10-08C0Calibration of TB-R (review bank)item windows, ICC, spend per cell12622.2 m$27.80
2026-10-10C0/TB-BTrial of three (kill switch). Calibration of TB-B (build bank)152.0 m$3.61
2026-10-10CA2Trial of three (kill switch). The setup's own version of the hook, skill-activation bank v2120.2 m$0.60
2026-10-10CA2The setup's own version of the hook, skill-activation bank v2+21.2 points at the interim look861.7 m$4.44
2026-10-10C0/TB-BTrial of three (kill switch). Calibration of TB-B (build bank)151.8 m$3.45
2026-10-10C0Calibration of TB-R (review bank)item windows, ICC, spend per cell12627.6 m$32.20
2026-10-11C0/TB-BCalibration of TB-B (build bank)v6 at ceiling: 31 of 37 items above 0.8510336.3 m$37.00
2026-10-11Screen 1One run per hard candidate, TB-B v713 of 16 hard greenfield slices at 1.02617.8 m$21.07
2026-10-12Screen 2ML/DL features and long-horizon reruns, TB-B v86 of 8 kept812.2 m$8.89
2026-10-12H1Harm check: the whole setup vs stock Claude Code, TB-B v9no harm; ×1.25 cost138105.2 m$93.78
2026-10-12H1 pass 2Both arms again on the 5 discordant itemsconfirms pass 11017.6 m$13.70
Total1,037260.1 m$268.80