What the ear heard and the instrument did not: a causal ablation of a reported memory exponent

A metric can be stable, reproducible, and measuring something other than what it is credited with. This note documents a causal ablation prompted by a blind listening comparison that contradicted the primary metric in use. Roughly eighty per cent of a spectral memory exponent reported for months as a property of a multiscale mechanism proved causally attributable to a single scripted periodic driver elsewhere in the rendering loop. An aperiodic control separated the contribution of periodicity from that of the driver's mere presence. A further finding, obtained by recomputation from values already in the record, is that this decomposition is not a property of the ablation but of the observation window at which it is computed: the share attributable to periodicity moves by more than thirty-five percentage points between the two windows present in the record. The hypothesis was formed after the observed failure rather than before it, and the work is recorded as a causal ablation rather than as a preregistered run. The note is published as a record of method rather than of result.