Scale Cuts and Nonlinear Modelling in Cosmic Shear: Where the Information and the Risk Reside

Survey analyses discard small-scale shear data whose modelling they do not trust, at a calculable cost in information. Integrating the signal-to-noise of the convergence spectrum for a representative sample, this paper finds that multipoles below 300 carry a fraction 0.238 of the total squared signal-to-noise, below 1000 a fraction 0.763, and below 3000 a fraction 0.983, so the discarded high-multipole region holds a substantial share of the raw statistical power. The large-scale primordial suppression studied in this set, by contrast, produces a fractional spectrum change of 0.016 at multipole 100 and only 0.0002 at multipole 1000: its signature lives where the information is thinnest and the modelling is safest. Scale policy therefore cannot be set by information or by safety alone; it must be set by where the specific signal under test resides. All quantities are computed from stated assumptions, not measured.