Weekly outcome ledger
Every dollar tied to work that shipped.
Agent activity reconciled against pull requests, CI, and review evidence—not tokens alone.
18 of 18 sessions reconciled · demo data
Cost per accepted PRPrimary outcome
$36.87 12% from prior week
Accepted PRsof 18
14+3 accepted outcomes
Acceptance rateTarget 70%
77%+8 pts week over week
Estimated time savedAccepted only
19.4hBased on owner estimates
Daily cost and accepted outcomes
Agent spend paired with pull requests that merged successfully.
Agent costAccepted PRs
$120$80$40$0
Mon 13
Tue 14
Wed 15
Thu 16
Fri 17
Sat 18
Sun 19
$516.18 total agent cost14 accepted · 2 rejected/reverted · 3 abandoned
Routing recommendation
Use Codex for routine test-fix tasks.
Across 11 comparable tasks, Codex matched the acceptance rate of Claude Opus with fewer retries and lower median spend.
42% lower
High confidence · 11 matched tasks
Latest sessions
5 of 18| Work produced | Agent model | Cost | Tests | Human repair | Outcome | Open evidence |
|---|---|---|---|---|---|---|
Claude Opus 4 | $18.40 | 83 passed | 2 corrections | Merged | ||
GPT-5.2 Codex | $7.20 | 46 passed | None | Merged | ||
Claude Opus 4 | $11.27 | 0 passed | None | Abandoned | ||
Claude Opus 4 | $42.80 | 118 passed | 3 corrections | Merged | ||
GPT-5.2 Codex | $31.50 | 64 passed | 1 corrections | Merged |