CDN-248 · The Calibration Diagnosis — GCFT Dark-Sector Book Arc (Art Edition)

CDN-248 executes CDN-247 directive (a): histogram CDN-242 S-class residuals against the CDN-246 v2 holdout S-class residuals to determine whether the CDN-243 S-band lower edge (0.10) is a calibration artefact. Verdict: CALIBRATION_ARTEFACT_CONFIRMED. The S-band lower edge exceeds 83% (5/6) of CDN-242's own S-class residual fractions and 100% (4/4) of the v2 S-class. CDN-242 S-class actual range: R/ξ_target ∈ [0.007, 0.134], mean 0.042; CDN-247 v2 S-class actual range: R/ξ_target ∈ [0.006, 0.061], mean 0.034. The 0.10 lower edge was an aspirational threshold set by judgment, not fitted to the empirical S-class distribution — it sits at the 95th percentile of the CDN-242 S-class. The proxy succeeds at minimizing S-class residuals, which is precisely what pushes them below the band. The CDN-247 CONFLICT verdict is driven entirely by this band-specification mismatch, not by proxy failure or genuine dark-sector tension. The CDN-242 population-averaged non-GCFT bound is UNAFFECTED (0.215% ≤ 0.29%, PASS). Zero single-cluster cap violations on either sample. Mean dark-sector closure 98.97% on the v2 holdout. The κ-proxy coefficients (b_refined of record from cdn242_residual_reduction_results.json) are NOT refit — CDN-247 lock respected. CDN-249 pre-registered directives: (a) propose revised S-band and P-band edges anchored to empirical residual distributions (percentile-based); (b) re-score CDN-247 under revised bands without proxy refit; (c) address the F-class super-ceiling absence (0 F-class clusters in v2 draw); (d) replace modelled cluster characteristic-parameter vectors with primary-literature pulls (carried from CDN-247 directive (c)). Art Edition with star-field design system (seed 248). Lock: CDN248_CALIBRATION_DIAGNOSIS_LOCKED.