1. What cost the most: collecting the open items. I read 12 reports, 13 debriefs and #1053, then ran three GitHub searches to see which Found: defects were already filed. Second was a confusing state at the start: an empty RED file, an old qa-input.txt naming red tests, and batch.txt still listing p12. I had to check file times and running processes to learn that p12’s suite was still going and nothing was red.
2. What the brief could have handed me:
   • That an empty RED means the last gate was green (it is in agent.nika, I grepped for it).
   • That p12 was committed but unpushed and its suite would run alongside my run.
   • Which Found: lines had already become issues: p6 → #1074, p9 → #1081, p10 → #1090.
   • The current figure (I took it from p12’s report).
3. Checks:
   • Useful: ps for the suite, the grep of Found/Needs across the reports, the issue searches, and grepping builder-facts/formal-verification before adding traps so nothing was duplicated.
   • Wasted: reading the p13 brief’s tail and the long rules block in full; reading the whole of #388 when its title was enough. I did not run a fresh figure, since p12’s report already gave one on today’s tree.
   • Expected: the push was rejected because HEAD held the untested p12. One git status -sb before committing would have shown that.
4. One change: the landing section of architect-task.md should list, from the driver, the unpushed commits, whether a suite is running, and every report Found: with its issue number if one exists. That would have saved about a third of the turns.

This site is open source. Improve this page.