Real coding sessions · never self-reported
One-shot session on adaptive-coach
2.8M tokens vs typical 414.6K
13h coding session using <synthetic>. 0 tokens across 1 prompts.
One-shot session on adaptive-coach
Previous best: 2.5M · +54%
13h coding session using <synthetic>. 0 tokens across 1 prompts.
One-shot session on adaptive-coach
Previous best: 2.2M · +14%
13h coding session using <synthetic>. 0 tokens across 1 prompts.
One-shot session on adaptive-coach
Welcome back
13h coding session using <synthetic>. 0 tokens across 1 prompts.
Debugged RE_SD
648 sessions tracked this week across all builders
87h 36m session running commands and testing. 159K tokens across 11 prompts using claude-sonnet-5-5.
Shipped evaluation validation framework with real metrics
First build tracked for this project
Merged held-out dataset validation system to main; all metrics now pass on real evaluation data.
Refreshed merge-accuracy metrics and benchmark baseline
First build using this model
Merged baseline refresh and metrics addition for more accurate merge-accuracy reporting.
Automated milestone tracking and fixed benchmark gate bug
First build using this model
Fixed collection timing bug blocking milestone 4, then automated roadmap status checks to run in CI.
Debugged RCC
1.2M tokens
45h 35m session running commands and testing. 1.1M tokens across 16 prompts using claude-haiku-4-5-20251001.
Verified anki management service is operational and stable
First build tracked for this project
Confirmed the anki management service runs stably.
Verified adaptive-coach cron setup and launchd configuration
First build using this model
Confirmed adaptive-coach uses crontab as documented, verified no launchd misconfiguration or orphaned processes.
Reviewed Pull-Day Training Plan Revision Designs
First build tracked for this project
Examined design specifications and planning documents for upcoming training plan modifications.