Published rejection lists reduce the variance ratio for paired and separate target resampling in three Kepler DR25 controls at 200–500 days

We tested how shared target stars change uncertainty in three public Kepler DR25 scrambled controls. We selected fitted periods of 200–500 days and measured multiple-event statistics of 7.1–12. We applied no stellar-type or radius filter. The raw tables contain 451 planet-candidate labels among 14,459 detections. We assume 198,640 independent target stars. Paired resampling gives each star the same weight in all runs. Separate resampling draws star weights independently for each run. The first-order paired variance is 1.195 times the separate variance. Its conditional percentile interval is 1.112–1.289 from 20,000 draws. A later exploratory check applies published rejection lists. It leaves 89 labels among 9,667 detections and reduces the variance ratio to 1.028, with interval 1.004–1.077. Thus, excess variance falls from 19.5% to 2.8% in this model. Shared labels, scores, and major flags agree with separate NASA outputs. Nine scrambled parameter rows are absent. Exhaustive membership bounds show that these omissions have little effect on either ratio. A separate parser checks the counts. Exact small-population resampling checks the formulas. We preserve source files, hashes, joint target counts, all draws, figures, code, and instructions. These results measure dependence in fixed controls. They do not establish a planet, isolate a physical noise source, or correct a published occurrence-rate posterior.

Kepler DR25 · scrambled controls · false alarms · reliability · target resampling · covariance · reproducible research