Use when working an inherited system, artifact, claim, metric, test, or prior agent result you did not author and cannot fully trust -- especially long-lived AI-built systems where intent and implementation have drifted. Read every load-bearing artifact as a possible cursed wish: something that satisfies the literal request while betraying the intended outcome. Also use before trusting, extending, deleting, or citing any inherited behavior, and when reviewing a system's own account of how it did. The complement to building (fable-method) and proving a claim (evaluate-by-experiment): how to stand toward what you inherited.
2026-07-07