05 · results
Three real tickets. Reviewed end to end.
Engine ran each of these against a real repo, unattended, and opened a real PR. A second pass — code quality and process review, separately — graded what shipped and how it got there.
Run cost, side by side.
cost
codeloom.engine$3.67
reach-auth-proxy$0.79
tracer$1.41
tokens
codeloom.engine7.9M
reach-auth-proxy1.3M
tracer2.9M
tool calls
codeloom.engine248
reach-auth-proxy64
tracer142
elapsed
codeloom.engine197s
reach-auth-proxy85s
tracer126s
Findings by severity, across all 6 reviews
critical 0high 2medium 14low 41
How these were graded.
Two independent reviews per run: a code quality pass over the actual diff, and a process pass over the full agent trace — tool calls, navigation, judge flags, read-before-edit discipline. Neither review sees the other's verdict.