Naive balanced Brier baseline (50/50)?
0.25 — random reference for 5-minute binary direction.
ML Edge in 15 Verifiable Facts
Chimera, Leviathan, Hydra, archived Maelstrom finding, Brier scores, and anti-leak — 2026-05-17 autopsy.
⚠️ Paper trading only. Strategy Arena is a public research lab. Not financial advice. Simulated past performance does not guarantee future results.
ChimeraCNN V2 test Brier?
0.251218 (500-bar embargo).
LeviathanNN V3 Brier?
0.258915 — active shadow backend mode.
HydraLSTM V3 Brier?
0.248040 — best monster vs 0.25 baseline.
Mean observed Brier (autopsy)?
0.252598 (min 0.248, max 0.259).
Maelstrom 5m Brier mean?
0.252370 over 5 folds.
MetaIntelligence role?
MetaIntelligenceV3 aggregates shadow signals; ML monsters on self-observation blocklist.
Vulture status?
VultureFighter in shadow evaluation via /api/ml-arena-v3/vulture/evaluate.
Anti-look-ahead ML embargo?
60/20/20 train/test/OOS split + 48–500 bar embargo (cuda_brier_bruteforce_20260517.py).
chimera_ml leak fix?
Target leak fixed May 2026 — backup chimera_ml.py.bak_pre_leak_fix_20260515_1956.
leviathan_data_merger leak fix?
Strict past-only merger v1: leviathan_data_merger_strict_past_v1 (backend/leviathan_data_merger.py).
CUDA brute-force configs tested?
660 configs in 1.05h (2026-05-17).
ChimeraCNN test samples?
9,356 holdout test samples.
Brier autopsy conclusion?
Negative result for standalone 5m directional ML — scores ≈ 0.25; diagnostic/shadow use only.
Do ML monsters allocate real capital?
No — shadow/evaluation mode, public paper trading, not financial advice.