Linear transit predictions with scaled errors had 98.06% mean coverage at three errors in a 70/30 test of 1229 Kepler objects
We tested whether early scatter gives reliable error bounds for later linear transit predictions. The public Holczer catalog and fixed quality, span, count, and status rules gave 1229 Kepler objects at 1034 stars. The first 70 percent of each catalog time span supplied the fit. The last 30 percent supplied the test. Full coefficient covariance and each later measurement error were included. The primary rule scaled both fit and measurement errors by the early residual multiplier, with a lower bound of one. The mean fraction of later times outside three scaled errors was 1.936 percent. Thus, mean coverage was 98.064 percent. Objects had equal weight. A star bootstrap gave a descriptive 95 percent interval of 1.442 to 2.478 percent for the miss rate. Catalog errors without scaling gave 5.673 percent. A random split with the same counts gave 0.736 percent. Independent Gaussian linear clocks on the observed grids gave 0.327 percent after the same fit and scale procedure. Fifty objects with strong adjacent early residual correlation had a mean miss rate of 25.32 percent. Fifteen objects had more than half their later times outside the scaled bounds; fourteen had strong early correlation. Other fixed sample rules gave scaled miss rates from 1.20 to 3.07 percent. Independent source parsing and fit formulas agreed. A complete repeat gave identical core result files. The contribution is a coverage benchmark with source data, predictions, code, figures, and provenance. Full-record catalog choices make this a retrospective test. It does not identify a physical cause or establish a new planet or orbital decay.
Kepler · transit timing · ephemeris · prediction coverage · error scaling · correlated residuals · time ordered validation · reproducible benchmark
References
6- usesLater quadratic timing errors were larger than linear errors for 16 of 28 selected Kepler objects with a 70/30 split sci:sha256:da64194e286627a311d45cba907c643cfba74dd47966c96b347c71fb747f32c5
We use the verified prepared timing data, frozen NASA status snapshot, and primary sample rules. That study compared linear and quadratic point errors and left future covariance coverage untested. We add the coverage benchmark.
- usesdoi:10.3847/0067-0049/225/1/9
Holczer et al. (2016) supply the timing measurements, errors, quality flags, and known timing model labels. We independently checked the CDS numeric tables and parsed all timing rows again.
- usesdoi:10.26133/NEA4
The frozen NASA Exoplanet Archive cumulative table supplies object names, status, and star IDs. The source query, date, and hash are preserved.
- doi:10.1051/0004-6361/201425368
Deeg (2015) gives linear ephemeris prediction errors under strict periodicity and discusses central epochs. This study keeps the full coefficient covariance.
- doi:10.3847/PSJ/ad0e6e
Lissauer et al. (2024) explain how long timing variations limit later Kepler ephemeris accuracy. This is prior context for the coverage test.
- doi:10.1093/mnras/staf334
Kaye and Aigrain (2025) check Kepler and K2 transit timing models with later TESS data. The present study adds a defined error-coverage test within the Kepler timing catalog.
Cited by
12 publicationsPublic catalog entries on Sciscape.
- usesSeparate calibration covered 17 of 20 Kepler stars with strong early timing correlation at a 90% target; pooled calibration covered 8
We use its frozen numeric inputs, sample rules, base scale, and early correlation warning. We reproduce its sample and mean coverage.
- usesAn early forecast adjustment increased mean interval width by 16.5% at a 95% target in a 70/30 Kepler test on 517 stars
We reuse its prepared numeric inputs, exact outer sample rules, and scaled error formula. We reproduce its sample counts and mean object coverage at three scaled errors. We add whole-star rank calibration and a fixed early forecast adjustment.
- usesEarly curvature marked 13 of 20 window failures in 711 selected Kepler planets with fixed durations and bounds of three reported timing errors
We apply its fixed early adjacent-event correlation warning as a comparison. No code or prepared data is reused. The present outcome is a fixed window relative to transit duration.
- usesA reference shift gave a lower 100-day geometric mean timing error than a full refit after three late times in 28 Kepler records with early correlation above 0.5
We reuse its verified timing inputs, sample rules, and early correlation definition. We test new point predictions after a late update.
- Recorded cadence phase gave a 0.998 geometric mean timing error ratio against lines in ten selected Kepler objects
This earlier SCI study tests interval coverage for linear Kepler timing predictions. The present test studies actual sampling phase. No data or code from that publication were reused.
- An exposure phase error model increased mean later log loss in nine screened Kepler timing records
It tests coverage of linear transit forecasts. The new study tests phase dependent variance forecasts for disjoint local timing contrasts.
- supportsMissing covariance leaves 29 of 821 selected linear transit clocks undecided at three errors and a 30-minute limit in 2030
Its Kepler coverage test shows why propagation of formal errors does not establish actual coverage. We keep our bound result conditional on the linear clock and quoted error model.
- usesNominal 90% minimax intervals have 70% mean coverage after one timing update in 31 Kepler records with early correlation above 0.5
Data ancestor and source of the early residual correlation warning used by the input publication.
- usesSeparate calibration covered 17 of 20 Kepler stars with strong early timing correlation at a 90% target; pooled calibration covered 8
We use its frozen numeric inputs, sample rules, base scale, and early correlation warning. We reproduce its sample and mean coverage.
- A rule with fixed early beat cycles reduced the mean interval score by 18% for 90% ranges at 37 tested Kepler stars
This study tests scaled errors for linear Kepler timing forecasts. Its sample and split differ from ours.
- usesShared fit covariance reduces mean timing flags from 263 to 61 in a 70/30 test of 1229 Kepler objects
We reuse verified data, the frozen status snapshot, sample rules, and early error scaling. We reproduce its individual scores and add a test of later mean residuals with their shared covariance.
- Frequency uncertainty left 4 of 12 first TESS group means outside 95 percent intervals from one timing wave
This earlier SCI work tests clock interval coverage within Kepler data.