Finite shifts reduce position bias in tests with fixed TOI-2095 TESS images

We test position recovery in empirical reference images from 14 TESS sectors of TOI-2095. Transit times, image weights, and temporal fits are fixed. In noiseless tests, quintic image shifts are fitted with a cubic model. Across three reference sets and nine imposed positions, the maximum position error falls from 4.01 arcseconds with pixel gradients to 1.41 arcseconds with spline derivatives and 0.16 arcsecond with finite shifts. The shifts are zero or axial offsets of 5 and 10 arcseconds. We also add 600 ppm loss images to full measured noise from nearby time controls. For centered 600 ppm candidate signals in 36 measured control groups, the finite-fit RMS position error is 4.80 arcseconds. The controls overlap. These recovery errors are not confidence limits. The real candidate fit uses 13 events. Its east and north shifts are 0.79 and 0.91 arcsecond relative to its reference image. Fixing both shifts at zero increases the weighted fit score by 0.72. These scores have no assigned detection probabilities. The tests measure recovery within fixed empirical image families. They do not calibrate the physical detector response or validate a planet. The package preserves all input arrays, source hashes, fitted images, control decisions, recovery results, figures, code, and repeat instructions.

TESS · TOI-2095 · TOI-2095.03 · transit source position · difference images · image interpolation · position bias · measured noise · reproducible research