Season offsets reduce Q10–Q17 depth errors for 15 of 19 Kepler signals selected from Q2–Q9 catalog data

We test whether early seasonal changes in Kepler transit depth repeat in later quarters. We fit a constant, four season means, and a linear time trend to Q2–Q9 quarter medians from the Holczer et al. catalog. We score Q10–Q17 with mean absolute error (MAE). Fixed coverage and quality rules retain 189 signals at 181 hosts. An early F-test screen selects 19 signals. Season means beat the constant for 15 of these 19. The median seasonal-to-constant MAE ratio is 0.445, with a host-bootstrap 95% interval of 0.377–0.617. Season means also beat the linear trend for 18 of 19. The correct quarter labels give lower errors than all three shifted-label alternatives in 16 cases. Across all 189 signals, only 82 improve and the median ratio is 1.033 (interval 0.997–1.078). Known seasonal depth effects are prior work. The new contribution is a fixed chronological benchmark with complete inputs, selection records, code, results, and figures. The source measurements use full-mission templates and flags. The result concerns recurrence in published catalog values, not a blind raw-flux forecast or a new planet.

Kepler · transit depth · seasonal variation · chronological validation · public astronomical data · reproducible research