Symmetric simulated eclipses give an apparent loss of asymmetry in a DI Herculis template test

We tested an empirical eclipse-shape method with four public TESS secondary eclipses of DI Herculis from 2020 and 2024. The two 2020 eclipses define an odd template and also receive fitted template coefficients. Each odd coefficient is divided by its fitted depth scale. The mean coefficient in 2024 minus the mean in 2020 is -1.007. All 72 tested analysis choices give a negative contrast. But 100 simulations with symmetric true eclipses give a median contrast of -0.894 and a central 95% range from -1.230 to -0.584. This range contains the observed value and overlaps the range for a source whose asymmetry disappears. A simple projection identity explains the bias in a common-grid limit. Independent-eclipse tests give contrasts of -0.173 and -0.667 for the two reference directions. The observed template has more odd power than the symmetric white-noise controls, so these controls do not show that the real eclipses are symmetric. None of 200 common-shape draws at each of three block lengths reached the observed decrease. The contribution is a numerical failure test for the specified estimator, with raw data, provenance, source code, figures, and reproduction instructions. The general data reuse problem is known. A physical shape change remains open in this study.

DI Herculis · TESS · eclipsing binaries · gravity darkening · data reuse · eclipse asymmetry · source controls · reproducible research