Experimental data: Titration of the Generational Memory Window - Engine and Brake in the Evolutionary Substrate

Data supporting the manuscript 'Titration of the Generational Memory Window: Engine and Brake in the Evolutionary Substrate'. (1) evolution_titration_pilot.json: v1 pilot (21 runs, global-truncation brake, all three pre-registered predictions FAIL, retained for the record). (2) evolution_titration_v2.json: v2 main experiment (40 runs, localbrake/globalbrake/nobrake arms, 5 seeds, 10^4 generations, per-generation trajectories). (3) evolution_titration_v2b.json: densified 14-point q grid (70 runs) with the frozen logit fit (pooled q_c=0.556, per-seed 0.523+/-0.008, R^2=0.977). (4) tau_probe_8000_results.json / tau_sweep_8000_results.json: SNN cross-substrate anchor and tau sweep at N=8000 (tau in {63k} and {40k/90k/126k}, 3 seeds, 300k steps). All pre-registration texts are frozen in the headers of the included scripts (evolution_titration_{pilot,v2,v2b}.py, tau_{probe,sweep}_8000.py); the SNN protocol body is ablation_gpu.py; figures are regenerated from the frozen JSONs by generate_figures_evolution.py without recomputation.