reversing an admission instead of deleting it

A candidate had been admitted into a collection on the strength of a strong measured result. Later, an unrelated defect fix changed how one component of that measurement was computed. Re-measured, the candidate did not merely weaken: the sign inverted. The quantity that had carried it into the collection was an artefact of the defect.

The instinct in that situation is to delete the record and move on. The record instead gained an explicit reversal entry — what was admitted, on what basis, what changed, and what the re-measurement says. Nothing was erased, so the history shows a collection that was once wrong, rather than a collection that was always smaller.

That cost far more than bookkeeping. Removing a member changes the combined result of everything that remains, so every survivor had to be re-adjudicated against the new baseline instead of being assumed to still pass. It also emerged that the weight attributed to the collection's constituents had been dominated by the withdrawn member; once it left, the remaining members stopped being distinguishable from one another and had to be re-weighted as a group. Two further members left for a different reason entirely — long-horizon behaviour that consumed capacity, and a variant with no protective boundary — and those were recorded as allocation decisions, not as defects.

What it proves: the programme can lose a result it had already adopted, say so in the record, and pay the full cost of redoing the work that depended on it.