an excuse tested before it was used

Months of results carried a caveat that had not been checked. A stretch of the input in one season looked thinner than the rest, and thin input is a comfortable explanation for anything measured over that stretch: it explains a bad result without requiring anyone to change their mind, and it costs nothing to leave standing.

The caveat was treated as a claim and put through a census instead. Volume by month, the shape of the per-minute distribution, the size of the peaks, the frequency of bursts, all compared across seasons and across years. If the season really had been sampled more thinly, the difference would have been visible in every one of those measures.

It was not. The autumn months were indistinguishable from the spring months on all of them. The only genuinely sparse input was one that had been sparse all year and was never part of the set being measured, which meant the caveat had been borrowed from a neighbouring question and applied where it did not belong.

So it was withdrawn. The record now says, in the same place, that the caveat was asserted, what test it was given, and that the test refuted it. The results that had carried it were re-graded without it, and the run under the corrected reading came back with the same null verdict it would have had to report anyway.

That last part is the interesting one. Withdrawing the excuse did not change the conclusion; it removed the reason anyone might have given for not believing it.

What it proves is that the programme audits its hedges as strictly as its claims.