Diminishing-returns stopping point

A stopping rule that ends iteration when the expected gain from another pass no longer justifies its cost, delay, or regression risk.

Later passes may add little, cycle between alternatives, or damage a previously good property. The stopping point depends on acceptance criteria, consequence, evaluator noise, and the value of improvement. A cosmetic rewrite can be waste on a factual report and the entire goal of a style task.

Builder example

A loop needs an external limit and observable progress measure. The model's own claim that changes are cosmetic is not an independent judge. Use task checks, score confidence, pass budget, repeated-state detection, or marginal outcome improvement, and route unresolved high-consequence failure instead of shipping merely because the cap was reached.

Common confusion: Diminishing returns does not mean the artifact is acceptable. It can mean the current method has stalled and needs new evidence, a different tool, escalation, or abandonment.