The NINJA suite employs cosmological hydrodynamical simulations to model galaxy populations at 5 ≤ z ≤ 10, using a fiducial run that varies spectral synthesis and dust attenuation prescriptions to match observed UV luminosity functions. Methodology relies on standard hydro codes with subgrid physics for star formation and feedback, though exact volume and particle counts are not specified in the preprint. Results indicate the dust-to-metal ratio evolves with redshift but normalizes differently by a factor of ~7 across models, producing large scatter in B-band, Hα, UV slope, and stellar-nebular color relations. This preprint, posted May 2026, remains unpeer-reviewed. It correctly flags non-convergence at z > 10 and underprediction of the UVLF even with top-heavy IMF assumptions, yet underplays how these dust variations interact with AGN feedback seen in complementary runs from the Thesan and Astraeus projects. JWST CEERS and JADES data (Finkelstein et al. 2024, ApJ) already show steeper faint-end slopes than many models predict, a tension NINJA partially reproduces but does not resolve without ALMA continuum constraints on cold dust. Mainstream coverage often ignores these quantitative population statistics, focusing instead on individual bright sources while missing the systematic model scatter that could reconcile or exacerbate apparent overabundances at z ≥ 10.