When the number is right and the claim is wrong: an adversarial re-examination of a distributed coordination benchmark

A quantitative performance claim can be a valid measurement and still fail to support the conclusion drawn from it. This note documents an adversarial re-examination of the author's own strongest quantitative result, conducted under a protocol written, digested and time-stamped before any comparison was run. Four findings are reported, all adverse to the author's position and all retained: the reported advantage did not survive comparison against a grid of calibrations of a conventional alternative; a declared system property supported by a passing test proved false at protocol level; the canonical figure was not stable across execution contexts; and the benchmark design did not support the distinctions drawn from it. The note is published as a record of method rather than of result.