95% bounds require unequal contrasts in 14 of 355 Gaia DR3 and SMASS II asteroid pairs under Gaussian errors with any dependence

We test whether unknown error dependence can remove differences between one fixed spectral contrast in Gaia DR3 and SMASS II. The sample has 355 numbered asteroids with Bus classes B, C, Cb, Cg, Ch, or Cgh. We assume centred Gaussian marginal errors with the listed fixed standard deviations and non-negative true reflectance. We permit any joint error dependence across wavelengths, surveys, and asteroids. A union bound over the fixed 396-object input set gives simultaneous 95% native-sample intervals. Positive interpolation weights give exact bounds for the nonlinear contrast. Fourteen pairs have separate intervals. All have a larger Gaia contrast. On the same coverage event, the range of the true survey differences has a lower bound of 2.275 percentage points after rounding. Thus, one common additive contrast offset cannot explain all pairs under this model. With twice the listed errors, two pairs remain separate; with three times the errors, none remain separate. A separate 44-nm ground-spectrum averaging check leaves ten separate pairs among 252 objects. We reproduce the earlier count of 147 pairs above three formal errors when covariance is set to zero. The new bounds remove that covariance assumption and control the primary sample at once. They do not establish a cause, new mineral absorption, or surface change. We preserve code, object-level bounds, source requests and hashes, figures, and instructions for reproduction.

asteroids · Gaia DR3 · SMASS II · reflectance spectra · measurement errors · unknown dependence · simultaneous confidence bounds · survey calibration · reproducible research