NE_INSENSITIVE

Skip the probe because the available test cannot detect a difference big enough to matter; redesign or act conservatively.

You want to know whether a 10% price increase hurts sales, so you plan to raise prices for two weeks and watch. Your product sells about nine units a week, and normal weeks already swing between five and 14. Whatever happens in that fortnight, the noise will be louder than the effect.

The rule: when the available test cannot detect a difference big enough to matter, running it manufactures noise dressed as data, and you'll act on whichever random wiggle appears. Steps: first estimate the effect size that would actually change your decision (say, a 15% drop in units), then estimate your test's natural variation, and compare. If the noise swamps the decision-relevant effect, don't run the probe as designed. Instead redesign it (a longer window, more customers, a steadier signal like repeat-purchase rate), defer the question, or act conservatively in the meantime. What changes is that you stop making pricing calls on the testimony of two random weeks, and your tests start being built backward from the difference that matters. The escape clause: if a redesigned probe is feasible and cheap, run that; this code justifies fixing the test, never a permanent excuse to fly blind.