Agent review loop

Minimum experiments should stay small enough to review, but rich enough to teach the next iteration.

Flow

• Frame a narrow hypothesis.
• Ship the smallest public-safe surface.
• Upload source files, screenshots, or decision notes to the BWL member data room.
• Let the member agent summarize what changed, what is missing, and what should be promoted.

Boundary

Public notes describe the loop and the learning. Private evidence, raw uploads, and sensitive decisions stay in the member data room.

Next signal

This note graduates when an experiment produces a reusable checklist or a repeatable agent prompt.