fix(supabase): remove non-existent columns from cycle INSERT

brain[cycle97]: restore app.py + 2x isotonic_temporal weight (7→14 init, 8→15 mutate; MDPI2026 ~0.199 Brier)

brain[cycle97]: restore app.py + 2x isotonic_temporal calibration weight (7→14)

brain[cycle97]: restore app.py + 2x isotonic_temporal calibration weight (7→14)

brain[cycle97]: 2x isotonic_temporal calibration weight (7→14, MDPI2026 ~0.199 Brier for LR+isotonic)

Brain cycle 82: fix calibration mutation bias — venn_abers boosted across all 3 GA impls

brain cycle79: space_id logging + cycle stagnation + hard reset

brain cycle79: lightgbm_brier model type + calibration mutation fix + S14 boost

brain-cycle-74: add isotonic_temporal calibration to GA

feat: MLP meta-learner for stacking (was LR, -0.003 Brier expected)

feat: add /api/best endpoint + feature_indices in to_dict

feat: calibration map + predict_today improvements + app.py enhancements

feat: add logistic_regression model type to GA (MDPI 2026: best tabular Brier 0.199)

brain: add lr_platt post-hoc Platt scaling calibration option (cycle 2026-04-01)

feat: add lr_meta calibration option (LR Platt scaling on temporal hold-out slice)

brain: add lr_platt post-hoc Platt scaling calibration option (cycle 2026-04-01)

feat: add lr_meta calibration option (LR Platt scaling on temporal hold-out slice)

feat: add lr_meta calibration option (LR Platt scaling on temporal hold-out slice)

fix: remove corrupted [Resource] prefix from app.py

feat: probability clipping [0.025,0.975] + LR as 14th model type

feat: MOVDA raw delta features (Cat37 6→13) + beta calibration + browser scraper

fix: LightGBM DART mode (boosting_type=dart, drop_rate=0.1)

fix: NSGA-II parsimony pressure + Venn-Abers calibration boost

feat: feature cap + rank fusion + TabICLv2 notebook

fix: cap adaptive mutation at 0.15 (data shows >=0.20 is destructive)

perf: remove stacking from CPU model types (best=0.24733, 10% worse)

fix: eliminate 40% wasted compute on CPU — neural models → tree-based islands

fix: xgboost_brier objective signature for XGBoost >= 2.0

fix: calibration hurts Brier — bias toward none + add Venn-Abers

feat: add xgboost_brier model type with custom Brier objective

perf: optimize evolution for CPU HF Spaces

feat: multi-island evolution + Codespace runner

feat: isotonic calibration + S10 optimal config (proven by 1244 experiments)

feat: unify feature engine + add version tracking to experiments

fix(stacking): CUDA crash + 4-model ensemble + tuned GA params

feat(adam): fix GA stagnation in app.py + 20 new features for S10

feat: scale GA to 500 population — island model, NSGA-II, 13 model types

feat(evolution_tuner): app.py — auto-generated improvement

feat(evolution_tuner): app.py — auto-generated improvement

feat(evolution_tuner): app.py — auto-generated improvement

fix: load game data from GitHub instead of nonexistent Supabase tables

feat(evolution_tuner): app.py — auto-generated improvement

feat(evolution_tuner): app.py — auto-generated improvement

feat(evolution_tuner): app.py — auto-generated improvement

fix: 5 critical Brier improvements — purge gap, calibration, regularization

fix: register experiment endpoints before Gradio mount on S11

feat: S11 experiment runner — isolated evaluation server for agent-proposed experiments

feat: add MLP neural network to GA model options

fix: replace circular ROI with log-loss proxy in fitness function

feat: NBA Quant Full Blast — 4 workstreams