The gated workflow - RapidReview

Merv’s premise: an agent doing open-ended research needs structure more than it needs freedom. So the unit of work is an experiment with hard gates, and the unit of learning is a reflection wave over many experiments.

Experiment level

Plan → Design review → Execute → Results review → Complete, with rejections looping back.

Plan → Design review → Execute → Results review → Complete, with rejections looping back.

Stage	Who	Gate
Plan	the research agent	writes a plan that states the claim being tested and how the result will be judged
Design review	a separate reviewer agent	approves, or returns to Plan — repeats until approved
Execute	the research agent	runs the experiment (locally or in a sandbox) and submits a report with evidence
Results review	a different reviewer agent	approves, returns to Execute (fix the run or the report), or returns to Plan (the experiment was flawed)
Complete	—	the experiment and its evidence become part of the project record

Reviewers are adversarial by design: their job is to find the reason the plan can’t test its claim, or the reason the results don’t support the conclusion. The brain enforces the gates — an agent can’t mark an experiment complete without an approved review on record.

Project level

Completed experiments fan out to three core lenses plus two written for the wave, then Synthesis, Reflection review, Publish.

Completed experiments fan out to three core lenses plus two written for the wave, then Synthesis, Reflection review, Publish.

A reflection wave snapshots every terminal experiment and every claim; three fixed lenses plus two written for the wave analyze it independently; the synthesis is reviewed before it’s published. Reflection waves

Why gates and not guidelines

A reviewer that can only advise gets ignored under time pressure. A gate can’t be.
Separating planner, executor, and reviewer into different agents removes the “grading your own homework” failure.
Recording the review chain makes every conclusion in the project traceable to the evidence and the reviewer who accepted it.

Details: Experiments and reviews · Reflection waves.

⌘I