Separate calibration covered 17 of 20 Kepler stars with strong early timing correlation at a 90% target; pooled calibration covered 8

We tested whether pooled calibration hides failures in Kepler stars with strong early timing correlation. We used public Holczer timing data and a frozen NASA status table. Fixed rules selected 1,229 objects at 1,034 stars. Lines fitted to the first 70% of each time span predicted the last 30%. One score measured the largest later error at each star. Separate sets of 517 stars supplied calibration and evaluation. A prior early residual correlation rule flagged 20 evaluation stars. At a 90% target for complete later star records, pooled rank calibration covered 472 of 517 stars overall but only 8 of the 20 flagged stars. Calibration within the two fixed groups covered 476 overall and 17 flagged stars. Flagged-group mean half-width increased from 226.4 to 610.7 minutes. Other-group mean half-width fell from 79.45 to 75.90 minutes. Overall mean half-width increased by 13.5%. One hundred other star splits gave median flagged-group coverage of 50.0% with pooled calibration and 94.44% with separate calibration at 90%. At the secondary 95% target, 10 splits had too few flagged calibration stars for a finite range under the rank rule. Independent source parsing and fit formulas agreed. The package preserves data, provenance, code, predictions, controls, and figures. This is a retrospective catalog test. The warning came from prior work on the same public data; it is not independently confirmed here. The result does not establish a new planet or a present-day transit range.

Kepler · transit timing · prediction intervals · group calibration · conformal prediction · retrospective forecast · coverage