CDN-246 · Disjoint Holdout v2 — GCFT Dark-Sector Book Arc (Art Edition)

CDN-246 re-drafts the holdout cluster sample after the CDN-245 disjointness audit ruled CDN-244 invalid. The pool combines the five confirmed-disjoint survivors from CDN-244 (Abell 2390, RXC J2228.6+2036, Abell 370, ZwCl 1358+62, Abell 851) with ten newly sourced clusters from CCCP-extension (Hoekstra+ 2015) and Weighing-the-Giants (Applegate+ 2014, von der Linden+ 2014): Abell 68, Abell 115, Abell 267, Abell 521, Abell 586, Abell 1413, Abell 1914, Abell 2219, Abell 2631, and MS 0451.6-0305. The audit gatekeeper (cdn245_audit.py) was run against cdn246_holdout_v2_inputs.json before the lock was issued; it returned PROCEED with 0 exact-name overlaps, 0 near-match overlaps, and N_disjoint = 15, satisfying the CDN-243 §3 minimum exactly. The audit script is now permanently embedded in the build step — the lesson from the CDN-244 → CDN-245 sequence is encoded in the toolchain, not in human discipline. All value fields in the input schema remain null. This paper locks the sample, not the data. The CDN-247 execution paper will populate values from the named literature sources, record canonical-JSON SHA-256 of the populated file, run cdn242_residual_reduction.py, and score the per-cluster R/M̄ landings against the CDN-243 prediction bands and decision rule. What carries forward bit-identically: CDN-240 locked physics block (ξ_target = 5.836, ξ_gal = 5.851, S_eff = 0.0623, ε_sat = 0.068); CDN-242 refined seven-term κ-proxy (b₆ = +21.98 at ln λ₀ = ln 17.5); CDN-243 prediction bands (S ±0.10, F ±0.20, P ±0.30, E ±0.05) and PASS/CONFLICT/FAIL decision rule; CDN-244 provenance schema (per-cluster source paper/table for every input, substitution_reason field, SHA-256 discipline). Art Edition with star-field design system (seed 246). Lock: CDN246_DISJOINT_HOLDOUT_V2_LOCKED.