A registered public-data reproducibility benchmark for the reported PKiKP detection of a Martian inner core

Artus Krohn-Grimberghe 

Bi et al. (2025) infer a 600-km Martian solid inner core partly from a PKiKP vespagram peak at 604 ± 2 s after P and −6.5 ± 0.6 s/deg from stacked InSight marsquakes. We present a registered public-data reimplementation; success criteria, windows, and statistical rules were frozen before outcome inspection, and the headline lane uses the public MPS deglitch stage. We recover a local maximum inside the published target box (601.95 s, −6.67 s/deg), but it never wins: the registered argmax is a displaced ridge (663.8 s, −3.64 s/deg) at 1.21× the target-box power, and six single-event removals each hand the 23-event stack's argmax to a time-adjacent shallow-slowness branch. Composition–outcome association gives p = 0.0039 and 1.0 × 10⁻⁴ in two same-data designs. Distance scrambles exceed ridge power in 75.5% of realizations and target-box-maximum power in 48.0%. A same-path α = 0 control exactly reproduces 4,600,000 cells; non-attested exploratory
injections first win at α = 0.25 in processed-trace pre-P-RMS units. The authors' raw-data deglitch method on the executable 19-event sub-stack preserves the qualitative disagreement and target-box subordination (box maximum cell-identical) while the global-argmax coordinate is deglitch-sensitive. This is reproducible non-agreement plus internal fragility, not a refutation; artifacts are released hash-pinned.