CDN-242 · The Residual Reduction — GCFT Dark-Sector Book Arc (Art Edition)

CDN-242 executes the three avenues pre-registered by CDN-241 on the validated 25-cluster sample, reusing the CDN-240 locked physics bit-identically. A refined seven-term κ-proxy with an intermediate-λ curvature term (b₆ = +21.98, centred on ln λ₀ = ln 17.5) lowers the mean proxy error from 0.041 to 0.029, cuts the dominant Class-S (sterile reservoir) residual mass by 43.3%, reduces the total residual budget by 16.9% (30.93 → 25.72 M̄), and raises mean dark-sector closure from 96.6% to 97.2%. The same refinement does NOT reduce the Class-P collisionless-offset residual — it rises (5.16 → 7.83) — confirming that residual as physical rather than a proxy artefact. A formation-history proxy φ cleanly separates the two Class-F clusters (φ̄ = 3.58, both super-ceiling with κ_req > 1) from the equal-κ_req Class-S interloper CL 0024+17 (φ = 0), establishing Class F as a genuine super-ceiling relaxed population. The surviving Class-P residual bounds the non-GCFT (particle-like) dark-sector fraction to ≤ 0.29% population-averaged (≤ 10.1% in the most extreme single offset merger); the GCFT coherence floor is ≥ 99.71%. The dark-sector residual therefore splits into a refinable part (S, proxy-resolution noise) and an irreducible part (P physical offset; F super-ceiling). Art Edition with star-field design system (seed 242). Lock: CDN242_RESIDUAL_REDUCTION_LOCKED.