A linear model of early dimming gives more total squared forecast error than a past mean for 18 selected TESS transits of BD+05 4868 Ab

We test whether early dimming predicts later dimming in the same transit of the known dusty planet BD+05 4868 Ab. Fixed quality and coverage rules select 28 TESS-SPOC SAP events from sectors 55 and 82. A past quiet window sets the baseline. Early input spans phase -0.06 to 0.04. The later target spans 0.10 to 0.30. Each linear model trains only on earlier selected events in the same sector. After five training events per sector, 18 targets can be scored. The early model gives 1.037 times the total squared error of the mean of past targets. Its root mean squared error is 831 parts per million; the past mean gives 816. The sector error ratios are 1.119 and 0.773. A time-only model also improves the sector 82 score. Removing one late sector 55 score changes the pooled ratio to 0.895. Changes to the flux reduction can also change the sign of the small pooled gain. Independent arithmetic reproduces the results. Tests that change only later flux leave all earlier forecasts fixed. A specified artificial profile with fixed shape gives a strong forecast gain. This is a retrospective benchmark with a small selected sample. It does not prove that physical dust variations are independent or that useful forecasts are impossible. We preserve the original data, provenance, every selection and prediction, source code, figures, and reproduction instructions.

BD+05 4868 Ab · TESS · disintegrating planet · transit shape · forecast · dust · reproducible benchmark