GraphNPE Inference Yields Cored Profile in Boötes I at ρ150 = 0.36 × 10^8 M⊙ kpc^{-3}

GraphNPE forward-models stellar velocities, measurement errors, and selection functions drawn from hydrodynamic simulations. Applied to Boötes I, it accesses line-of-sight kurtosis to break the mass-anisotropy degeneracy that biases dispersion-only fits toward cusps. The resulting posterior places the galaxy among the lowest-density systems at its stellar mass, directly testing CDM predictions for ultra-faint dwarfs where baryonic feedback is minimal.

Literature values for Boötes I typically assume NFW profiles and report densities several times higher. GraphNPE's lower result aligns with the small-scale structure tensions seen in other dwarfs yet remains within the broad uncertainties allowed by current data. For Draco the same pipeline recovers cuspy inner slopes across independent datasets, underscoring that core formation is not universal at these masses.

The principal limitation is Boötes I's modest sample size, which leaves the inner slope only marginally constrained. Deeper spectroscopy providing several hundred additional member stars with precise velocities would tighten the kurtosis measurement and distinguish core radii of 100–300 pc from cusps at >3σ.

Future wide-field surveys such as DESI and Rubin LSST will supply the required samples for a statistical test of core incidence across the ultra-faint population.