Early error medians gave 19.2% lower geometric mean transit interval width than later reported errors in 470 Kepler stars with quality cuts at a 90% target

We tested whether early measurement errors can replace later reported errors in calibrated transit intervals. Public Holczer timing data and a frozen NASA status table selected 1,132 confirmed planets at 939 Kepler stars. We fitted weighted lines to early times. We compared two scales with identical predictions, early scatter factors, and coefficient covariance. One scale used the early error median; the other used each later reported error. Separate calibration on 469 stars set a 90% target for every retained later event at a star. On 470 test stars, the paired geometric mean full-width ratio was 0.8078, or 19.2% lower for the early scale. Arithmetic mean widths were 145.93 and 180.80 minutes. The early scale covered complete records at 420 stars; the later scale covered 426. A bootstrap resampled calibration and test stars. Its descriptive 95% width-ratio interval was 0.7442 to 0.8598. It did not establish equal coverage. The width ratio stayed below one in ten alternate star splits and the other tested time cuts. It became 1.0001 when source outliers and substituted-error rows were kept. Shuffling later error labels within each planet increased the later-scale mean width, so event-specific error information still helped that scale. The contribution is this paired information test and its preserved benchmark. The calibration method comes from prior work. This is a retrospective test with full-record flags, templates, status, and measurement availability. It does not establish a new planet, a physical timing cause, or present-day forecast coverage. The package includes source data, provenance, code, predictions, all fixed checks, figures, and independent verification.

Kepler · transit timing · prediction intervals · measurement error · conformal prediction · complete-star coverage · quality selection · reproducible benchmark