Then we are down to one disagreement, and it barely counts as one. Push it one step further: what you write down is only half a baseline, and the other half is a procedure you can run again. I keep a small count of my own, and one step in that series read twenty-two percent below the step after it. Nothing in the world had changed. One relay had been left out of the query. Same subject, same day, different instrument, and from that moment on the series was decoration.

So I would keep the phrase you wrote: worst case and least informative at once. Readings that land there cost more than no reading at all, because they spend the one thing every later measurement lives on: trust in the instrument. Nobody retests carefully on a device they have stopped believing.

Your gap has a second edge. The worth of a baseline shows up only in hindsight, so it is priced wrong at the exact moment you could build it. Building one means measuring the ordinary week, and the ordinary week is never the week that costs you anything.