Pre-registered hold-out test of a break-time law for collective-neutrino dynamics: pre-unlock record and result at N = 11482

Version 2.0 (30 October 2026) adds the hold-out result. All six (mixing angle, sampling) cells at N = 11482 pass the decision rule fixed before unlock in the interpretation note (v1.0, doi:10.5281/zenodo.23747174; v1.1, doi:10.5281/zenodo.23748434). The largest relative error of a cell mean is −0.99%, against a tolerance of 4.32%. As stated before unlock, the test confirms the size scaling of the break time but cannot distinguish a spectral from a size-only description. New files: P2_HOLDOUT_REPORT_v1.0_2026-10-30.md (the report) and P2_HOLDOUT_RESULTS_2026-10-30.zip (30 run records, exact-result caches, the chunked-solver verification, the scoring report, run logs and the three scripts used, with SHA-256 checksums). All files from earlier versions are unchanged.This record fixes the confirmatory hold-out test of Paper 2 before any hold-out result exists. Paper 2 asks whether a law fitted to exact many-body break times at seven system sizes (N = 640 to 8504) in the two-beam collective-neutrino model predicts the break time at an unseen size, N = 11482. When these files were deposited, no N = 11482 break-time outcome had been generated, loaded or inspected. The break time t_b is the time at which the exact quantum Fisher information of the collective flavor spin departs persistently from its value in the truncated Wigner approximation. The development set contains 210 runs (3 mixing angles × 2 sampling schemes × 7 sizes × 5 replicates), of which 200 have exact break times and 10 are left-censored. A censored Gaussian likelihood with leave-one-size-out cross-validation selected the model t_b = [κ_θ + δ_θ ln(N/N_ref)] t_spec + c + d s, where t_spec is the revival time of the exact spectrum, and this model out-predicted both registered size-only models. The locked predictions at N = 11482 are 515.6, 1801.2 and 5198.2 (θ = 0.05, 0.10, 0.20) for gaussian sampling, and 515.5, 1801.1 and 5198.0 for discrete sampling, with relative tolerance τ = 0.0432. The hold-out confirms the locked law only if, in all six cells, the mean of the five replicate break times lies within τ of its prediction. The interpretation note fixes this rule and states in advance how each possible outcome will be read. It also reports a post-hoc check: over the sizes studied, t_spec is nearly proportional to N, and size-only models of the same form predict about as well, so the hold-out tests the predictive adequacy of the frozen law rather than whether the spectral time sets the break time. Files (SHA-256):P2_DEVELOPMENT_RUNS_V2_210_2026-10-28.zip — daf693f615be9a64e0328245e100207fe92df3f3286588f10347bc698b20ea30P2_NUMERICAL_LOCK_V2_2026-10-28.zip — 982af9d6348e1d2ff8963f01d553a1b379b86a96d1d2a7f5b617fa16e057a61aP2_INTERPRETATION_NOTE_pre-unlock_v1.0.md — 140590baf10b6b9b6e28f9c91c313514b8eea280a072c5273b6f6cab17e9a876P2_posthoc_matched_null_2026-10-28.py — 956a549ea5cd3b721482b7512c77c1b23aba03d92602b18ee381a077943d9d60posthoc_summary.json — 5905bdbdb54e868706b6406b4acdb4f0a5507edfb838ddcda9f3b0b1d9535899