Model exposure sampling reverses radius and orbit rankings in two of eight selected circular Kepler injections at fixed impact parameter

We test whether exposure sampling changes a transit-shape comparison in public Kepler DR25 INJ1 data. Model values are averaged with equal weights, including both exposure endpoints. We select eight radius-mismatch targets and four controls by input cuts and a fixed hash order. One control fails the paired-flux check. A quarter-availability amendment is reported separately. At fixed listed impact parameter, two circular candidate models change either radius or orbit scale to match the same listed duration. Both have the same nominal contact times. We fit a physical quadratic limb profile, bounded timing changes, and an amplitude per event. Alternating cadences provide training and test samples. At 31, 101, 401, and 1,001 samples per exposure, all eight mismatch cases favor the radius change. Both 31-sample test splits give this ranking. A later source-motivated check uses 11 and 21 samples. KIC 9701755 and KIC 7466078 then favor the orbit-scale change. Their radius/orbit error ratios change from 4.531 and 1.433 at 11 samples to 0.034 and 0.053 at 1,001 samples. The other six mismatch cases retain the radius preference at every tested count. Thus, coarse exposure sampling reverses two selected geometry-model rankings while nominal contact times stay equal. The tested averages do not reconstruct production integration. The result does not establish a true input radius, a population rate, or a production cause. Source data, provenance, code, predictions, figures, and reproduction checks are preserved.

Kepler · DR25 · transit injection · transit shape · ingress · radius ratio · data validation · reproducible astronomy