A Calibration Audit of a Gaia XP White-Dwarf Main-Sequence Binary Catalog: How Much BP-Band Residual it Takes to Manufacture Contamination

Abstract

A Gaussian-process classifier on Gaia DR3 XP spectra produced $\sim$30{,}000 white-dwarf main-sequence (WD--MS) binary candidates, each with a probability but no likelihood or goodness-of-fit. The corrected XP BP band keeps a local blue-flux residual of about 2\%, where a hot white dwarf also adds flux. We asked whether that residual contaminates the selection at its realistic amplitude. It does not. The answer turns on one unit: 2\% of total flux, deposited in the narrow blue band, is a median 55\% local excess on a red MS star, 27 times the same number read locally. Injected as a 2\% local excess it gives a spurious rate of 0.08 on a 0.05 baseline through the $Δχ^2$ threshold standing in for the classifier, and an amortized posterior keeps 0.84 of its 90\% coverage against a clean 0.88. The selection fails only above a 10--20\% local excess and reaches 0.96 near 50\%, the raw uncorrected bias that correction and the $B<18$ cut remove. A model-comparison gate certifies binarity only above a WD flux share near 0.05; at the catalog&#39;s median share the statistics invert and the spurious carry the larger $Δχ^2$ improvement. The clean reliability signal is off-cooling-sequence UV deficiency (GALEX FUV detection 0.19 against 0.50), robust to a distance control. At the residual the correction leaves, the contamination hypothesis is a null; the open regime is the faint half, where that residual is unmeasured. The audit gives the amplitude it would take, and the shape of the failure past it.