What stood out to me in this article is the idea of AI exposing inference constraints rather than proposing new physics.
I ran into something similar working with the SPARC disk galaxy data. A simple acceleration-domain normalization stayed coherent across morphology bins and survived several attempts to break it (distance perturbations, alternate radius choices, surface-brightness splits), without any fitting.
What surprised me most is that it didn’t collapse in the dwarf / low-surface-brightness regime, which is where I expected systematics to dominate.
What stood out to me in this article is the idea of AI exposing inference constraints rather than proposing new physics.
I ran into something similar working with the SPARC disk galaxy data. A simple acceleration-domain normalization stayed coherent across morphology bins and survived several attempts to break it (distance perturbations, alternate radius choices, surface-brightness splits), without any fitting.
What surprised me most is that it didn’t collapse in the dwarf / low-surface-brightness regime, which is where I expected systematics to dominate.
Full analysis here (runs locally, CSV outputs): https://github.com/JasonResearch/sparc-coherence-discrepancy...