LLM refinement of decompiled code recovers names from the model's prior, not the decompiler output.

A within-item control costing twenty API calls: refine a function, then refine it again from an input whose identifiers have been destroyed, and measure what survives. Recovery is real (+0.072 to +0.137 above an arm-matched permutation null), but destroying the input's dataflow changed the naming gain by +0.001, CI [-0.026, +0.026]. A second refiner from another vendor, pre-registered with byte-identical inputs, reproduced it across twelve contrasts. Readability stayed at ceiling throughout, so a human reader gets no signal the names came from the prior rather than the binary.