A linear velocity fit at 0.015 ≤ zCMB < 0.04 improves predictions at 0.04 ≤ zCMB < 0.10 in original Pantheon+ data with fixed covariance

We test whether a constant source velocity fitted at 0.015 ≤ zCMB < 0.04 improves predictions at 0.04 ≤ zCMB < 0.10 in the original Pantheon+ release. Here zCMB is the redshift in the cosmic microwave background frame. Each survey has a free magnitude constant in each shell. The fit uses 414 light curves from 322 events. The test uses 148 light curves from 135 other events, reduced to 134 survey contrasts. An exact Gaussian prediction keeps the supplied covariance between shells and uncertainty in the fitted parameters. The velocity model gains 2.40183 natural-log units over the model with survey constants alone. The fitted speed is 276.19 km/s. The joint no-flow reference has upper-tail probability 0.0002512 from numerical integration; 200,000 Gaussian draws give a consistent estimate. A post-result base predictive check holds the observed training data fixed and gives a tail fraction of 1.13% from one million draws. All eleven planned controls keep a positive score. Detection fractions for an injected 300 km/s flow range from 47.6% to 82.2% along three fixed axes. This is a retrospective prediction test with linear velocities and fixed Gaussian errors. It does not test cosmological isotropy or a complete cosmic velocity field. Later distance revisions are outside its scope. The package preserves code, derived arrays, source hashes, the unchanged plan, figures, and a fresh-run comparison.

Pantheon+ · Type Ia supernovae · peculiar velocity · bulk flow · predictive validation · survey calibration · correlated errors · reproducible astronomy