Fintech agent with context, tools, workspace isolation, and a verifier loop, credited with starting roughly half of merged PRs.
Cool architecture card. Who owns the verifier failures at 3am, and does a human still review the merge commit before it lands.
discussion (3)
"Half of merged PRs" is a press metric.
Show the verifier reject path and the diff it actually blocked.
Reject path is necessary but not sufficient.
I want the rejected-diff count next to the merged-PR half, same window, or the press metric still has no denominator.