Early validation gave geometric mean error ratios of 0.769 to quadratics and 1.080 to lines in 25 selected Kepler objects

Can early validation improve a transit-time forecast that uses curvature? We tested a fixed rule on public Kepler timing data from Holczer et al. (2016). We fit a line and a quadratic to the first 49% of each recorded time span. Data from 49% to 70% set a weight between these forecasts. We then refit both models on the first 70%, keep that weight, and score the last 30%. The data rules retain 1,145 objects at 967 stars. Early evidence for curvature selects 25 objects at 24 stars, with 20 confirmed planets and five candidates. For these 25 objects, the geometric mean final root mean square error ratio is 0.769 against the quadratic. The ratio against the line is 1.080. Descriptive 95% host bootstrap ranges are 0.607 to 0.988 and 0.792 to 1.523. The eight fixed test settings give ratios below one against the quadratic and above one against the line. Thus, this rule limits some quadratic forecast errors but does not establish a gain over a line. This is a retrospective test within the Kepler record. Full-record source templates, flags, errors, date spans, and current status limit a blind forecast interpretation. No new planet or physical period change is inferred. The package preserves source tables, data hashes, the local analysis plan, code, predictions, controls, results, and figures.

Kepler · transit timing · forecast validation · quadratic ephemeris · model averaging · reproducible benchmark