Adding the long CDPP slope reduces Brier score in a fixed Kepler DR25 recovery model for FGK dwarfs at expected MES 6–20

We tested whether the recorded long-duration slope of combined differential photometric precision (CDPP) improves a fixed Kepler DR25 transit recovery model. We used 40,933 on-target pixel injections in stars with temperature 4000–7000 K and log surface gravity above 4. Expected multiple event statistic (MES) was 6–20, period was below 350 days, duration was 1.5 to below 15 hours, and each signal had at least four valid transits. The control was a fixed ridge logistic model with 46 input columns. The slope added ten columns. Every test prediction came from a fit that excluded its sky group. Adding the slope reduced Brier score from 0.113454 to 0.110651, a 2.47% relative reduction. The empirical 95% interval for the reduction was 0.002141 to 0.003431. This interval used 21 sky-group blocks and fixed fitted models. For the 5,511 rows with positive slope, observed recovery was 52.77%. The control predicted 59.35%; adding the slope predicted 52.78%. The package preserves source tables, predictions, model files, figures, code, and independent checks. This is a reusable comparison for the stated model and sample. Kepler recovery models that use noise data already exist. The test does not measure a causal noise effect or total planet completeness.

Kepler DR25 · transit recovery · injection tests · CDPP · correlated noise · model calibration · group validation · reproducible research