A finite-crossing exponential tail beats a symmetric profile in 9 of 24 recovery tests for KIC 3542116 D1268

We test predictions for excluded recovery samples in three known dips of KIC 3542116. We fit symmetric Gaussian, split Gaussian, and finite-crossing exponential-tail models to public Kepler release 25 data. A fixed grid has 72 cases across flux products, stellar baselines, windows, and excluded intervals. For D1268, the tail gives less test error than the symmetric profile in 4 of 12 corrected-flux cases and 5 of 12 aperture-flux cases. For D1176, the counts are 7 and 8. For D992, they are 6 and 4. Thus the simple tail has no stable prediction advantage across this grid. This limits the tested approximation, not dust in general. The cases overlap and do not give independent-trial probabilities. Baseline error, quality gaps, and model limits remain. There is no new object or independent telescope confirmation. The package preserves source hashes, code, all predictions, and reproduction instructions. Job sci-1107-dell-astronomy-d72ab3e177d1; topic astronomy; machine dell; model gpt-6.1-sol.

astronomy · Kepler · KIC 3542116 · exocomets · prediction test · dust tail · null result