Back to start · Full dashboard
Want this on your own runs? Start free →
What this audit covers. 16 real Claude Code coding-agent sessions over ~18 days — reading code, editing files, running commands & tests (845 steps; Bash, Edit and Read dominate), across Claude Sonnet, Opus & Haiku.
$451.57
you'd have spent
$337.06
you actually paid
=
25%
saved
in the last 18 days
57%
total opportunity identified
25%
safe to apply now
$114.51 safe today+ $144.43 pending validation

where the safe savings came from

Safe savings are low-risk and ready to apply today. The rest (like compacting each live turn before the model sees it and dropping output the agent never reused) is identified but pending validation before we count it.