Proverbs
Adversarial review
Every plan deserves a critic with no stake in it. The machine will play one, but only if you ask.
The tenth law scales past your own drafts: every plan that will spend other people's time deserves a critic with no stake in it, and the machine will play that critic well, but only if the review is staged honestly. A model reviewing the plan it helped write, inside the session that wrote it, is a defendant reviewing its own case. Fresh session, no authorship, explicit instructions to attack: that is a review.
For plans that matter, make the review a panel. One pass graded against your stated principles, one hunting for the constraint nobody mentioned, one asked bluntly why this fails. Grade against the rubric, not the intention; the reviewer who knows what you meant will find that you meant well. Then reconcile the passes yourself, because synthesis is the steward's half of the job.
How to practice it
Fresh eyes, literally.
Review happens in a session that didn't write the plan and owes it nothing. Authorship is a conflict of interest, in machines too.
Attack, don't admire.
The instruction is to find what fails, not to assess. Reviews scoped to assessment return compliments.
Grade against the rubric.
Score the artifact as written against the stated criteria, not against what it was trying to be.
You do the synthesis.
Panels produce conflicting findings. Reconciling them into decisions is judgment, and it stays with you.