A Reproducibility Benchmark for a CSL Force-Noise Calculator: Two Reproductions, One Unresolved Discrepancy

A source-bound benchmark is presented for a white, mass-proportional Continuous Spontaneous Localization (CSL) force-noise calculator. The calculation is compared with fixed numerical points from a levitated-sphere study, a nanocantilever study and a LISA Pathfinder analysis. After explicit spectral-convention reconciliation, the sphere comparison reproduces its five assigned source points with a maximum relative deviation of 8.66%. A rigid-cuboid-plus-sphere cantilever approximation reproduces five points from the published contour with a maximum deviation of 11.28%. Both satisfy the fixed 30% point criterion. The LISA calculation clears the smaller assigned points but fails at r_C = 0.1 m: computed/source is 2.03507. The supplied digitized trace also shows disagreement across approximately 0.03–0.30 m, with ratios 1.95–2.77; its onset remains unlocated. The completed labelled-tick calibration leaves a 30.31-pixel discrepancy against the fixed ±5 allowance. A portable kit regenerates the results, including this retained failure. These checks test implementation and source reproduction, not CSL’s physical validity, a new exclusion region or apparatus feasibility.

This first public release (v0.1) derives from adopted R4. The record includes the manuscript PDF and DOCX and a portable code/data kit. Original code is MIT licensed; original manuscript, narrative documentation and newly drawn figures are CC BY 4.0. Third-party source/dependency rights remain with their holders; see LICENSES.md in the kit.

Generative AI assisted planning, implementation, source checking and drafting. Same-session and related-family reviews do not establish independent validation. QianJun Yu is responsible for the release. The LISA failure is intentionally retained.