Principles

Eight working habits of the author. They are a reminder, not the basis of a review: that is the literature in References, and where the two differ the literature wins.

Principle	In development	In a review
First principles	Reproduce the problem and find its cause first	A blocking finding needs a failing test with an oracle
Adversarial review	Someone who did not build the change looks for counterexamples	Experts did not write the change, and only the verifier records test results
Ablation	A part whose removal changes nothing measurable is removed	An expert stays only if the evaluation shows it helps
Occam's razor	Build the smallest version that works	A path no review can reach is removed (Levels)
Uncertainty checklist	Each handoff lists what lacks evidence	Each report ends with absent experts, unreproduced findings and skipped runs
Independent thinking	Reach conclusions separately before comparing	Experts do not see each other's output
Critical thinking	Say what would disprove an explanation	Each finding states its evidence kind
High cohesion, low coupling	Draw boundaries so a part's context is easy to state	Each agent gets its files, its skills and the plan