registering the test before paying for it

Any interesting question in a long research programme can be asked a dozen ways. Ask it a dozen ways and one will look good; that is arithmetic, not discovery. The programme's answer costs a little time every single day: a measurement is specified, and the specification committed, before compute is charged.

Each registration names the comparison, the data era, the cost basis, which columns are primary and which are sensitivity, and — the part that matters — the rule under which the result counts as a reversal. Overturned if the conservative row clears the bar. Overturned if the qualifying rate exceeds one half. Written first, in the same document as the setup, then left alone.

The charged cell is recorded against that registration afterwards. A run cannot quietly become a different run. When a basis turned out to be wrong, the fix was not to reinterpret the old output but to register a new measurement and mark the old one provisional in public.

The cost is visible in the record: days where the honest answer was that the run had not started because the inputs did not exist yet, logged as a queue entry rather than reported as a result. Nothing gets announced twice.

It proves the team can be told a disappointing answer, because it has arranged to be unable to tell itself a comfortable one.