---
name: mutate-the-defect-not-the-trigger
description: "A mutation proof that varies the TRIGGER while the DEFENSE stands measures nothing — remove the defense, then re-apply the trigger."
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 7833f7bf-2717-4d3d-b443-fc94345eb5f6
  modified: 2026-08-04T14:31:43.989Z
---

A tooth-proof mutation must remove the DEFENSE the test guards, not merely re-apply the TRIGGER that once exposed the defect. Measured 2026-08-04 (U3 stack wrapper, todlando): putting 3 args back with the 16 MiB wrapper still installed let T5 pass — correct behavior, but the mutation measured nothing; taken at face value it read as "smoke unproven" and would have sent the builder hunting a fault in a sound test. The real kill (MUT-C2) reverted the wrapper AND restored the trigger: both smoke rows red with the defect's verbatim face.

**Why:** trigger and defect are different objects — the defect was "CLI runs on the main thread's stack", the trigger was "argument count". A defense sized far above the trigger absorbs any trigger-only mutation, so survival discriminates nothing.

**How to apply:** when a mutation survives, first ask whether it removed the defense or only varied the trigger. Only a defense-removing mutation's survival indicts the test. Kin: [[make-a-new-rig-red-on-purpose]], [[setup-suppresses-failing-arm]].