40 Kepler-9 scores imply unequal point likelihoods with independent Gaussian errors, 13 parameters, and 97 or 107 transit times

We test the statistical support of 40 published Kepler-9 orbit points. We preserve their numeric source rows and public transit timing records. We assume each reported reduced chi-square equals chi-square divided by N−13. All points use the same independent Gaussian errors. The catalog gives count scenarios N=97 and N=107. These counts convert the listed scores. We do not calculate the source scores again. For N=97, scores 1.65 and 2.13 give a point likelihood ratio of 1.76 × 10⁻⁹. A fit of one common error scale gives 4.18 × 10⁻⁶. Score rounding does not remove this difference. Also, 35 of 40 printed mass ratios fail a check that allows rounding of both masses and ratios. Separate local orbit fits use an explicit 97-record selection and three starts based on the listed values. They reach reduced chi-square values near 1.339 and masses near 44.2 and 30.4 Earth masses. They stop on small changes, not on the gradient test. The package preserves code, data, source hashes, fit status, numerical checks, figures, and reproduction instructions. The score audit is conditional on the score definition and noise model. Point likelihoods do not give mode probabilities, a unique global orbit solution, or new mass limits.

Kepler-9 · transit timing · planet mass · likelihood · reproducibility · table audit