bi-14 step 03 — The record, including rejections

Goal

Convert code review into durable learning — the part almost everyone skips.

Tasks

For every attempt, landed or not, record:

FieldNote
bugas the project sees it
reproductionthe first thing you lose if unwritten
specwhat should happen, implementation-independent
suspected subsystemyour prediction — score it
source path + call pathso the navigation transfers
testwhat would have caught this
proposed fixincluding alternatives you rejected
reviewer feedbackthe highest-value field
architectural lessonwhat you now believe that you did not
  1. Complete the record for each rung attempted.
  2. Write the architectural lesson even for rejected CLs. A rejected CL with a clear lesson is a successful lab.
  3. Score your subsystem predictions across all attempts. Compare with your bi-01 drill accuracy — has navigation calibration improved?

Done when

  • A record exists per attempt, rejections included
  • Prediction accuracy compared against bi-01
  • PROGRESS.md contribution ladder updated honestly