Pixel noise assumptions change the preferred source fit in eight TESS sectors of KIC 8164262

We test the spatial source of the known 2.61849265 cycles per day oscillation in eight public TESS sectors of KIC 8164262. The sample has 54,596 accepted cadences. We preserve 8,780 bins with all 121 pixels. One variable point source is fitted with a background plane and bright-star image terms. Diagonal pixel weights give competing fits near the target and the bright neighbor TIC 159649652. Their residual scores are 38.68 and 28.89, respectively. We then include a stated fraction of the full residual block covariance. At covariance fraction 0.9, the scores are 141.70 and 169.38, so their order changes. Absolute scores from different covariance models cannot be compared as likelihood evidence. The empirical covariance is singular, and the tested fractions are sensitivity controls. Added bright-star second derivatives, a bright-core mask, or removal of first derivatives also favor the target region. Exact catalog-position tests and routine local controls miss the displaced competing fit. Thus local precision does not resolve source identification under the tested model choices. We do not identify a new variable star, planet, or frequency change. The package contains offline pixel inputs, archive response files, source hashes, all tested models, simulations, figures, code, and reproduction instructions. A replay of all eight temporal measurements gave zero numerical differences in the recorded environment.

TESS · KIC 8164262 · TIC 159649612 · TIC 159649652 · source localization · pixel response function · correlated noise · covariance regularization · heartbeat binary · tidal oscillation · public astronomical data · reproducible analysis