Methodology & Transparency
| Domain | Evidence | Gate | Status |
|---|---|---|---|
| Live strategies | arena_state_*_v5.json | Public count | 75 |
| ML / Brier | brier_autopsy_*.json | anti-leak OOF | published |
| Monte Carlo CV | live_mc_results_snapshot.json | Sharpe_p5 | 7 |
| Strategy Hospital | JSON | triage | PASS/WATCHLIST |
| Mode | paper trading | No real orders | paper-trading |
How Strategy Arena's ML and statistical layers actually work. The Brier scores published here are dated measurements; an expected value is marked as an estimate. The fixed leaks are those found by the 2026-05-15 audit, listed below.
Anti-marketing: if a layer is analytics, we call it analytics. If it is rules-based, we call it rules-based. If it is ML, we publish its measured Brier when one exists; otherwise the value is marked as an estimate.
Atlas Edge Allocator · Live MC Results · ML Edge Report · Portfolio MC · Strategy Lifecycle · Edge Radar
Atlas Edge Allocator · Live MC Results · ML Edge Report · Portfolio MC · Strategy Lifecycle · Edge Radar
New: Strategy Hospital publishes live strategy triage, and Strategy Lifecycle keeps the public history of those changes.
The 4 active monsters + 1 archived finding
| Monster | Architecture | Metric (measurement, estimate or archive) | Status |
|---|---|---|---|
| Invictus ML Ultimate | LightGBM with isotonic calibration, OOF validation and monotonic constraints | Brier OOS expected ~0.22 Estimation expected value, not a published measurement |
Real ML Audited 8/10 by DeepSeek |
| Chimera Scanner + CNN | 17 statistical patterns + PyTorch CNN, 108 OHLCV/pattern channels | Brier OOS 0.2512 9,356 samples Measurement brier_autopsy_20260517.json, generated 2026-05-16 |
Hybrid Rules + real ML |
| Leviathan 9-Layer Ensemble | 8 heuristic layers + 1 PyTorch MLP as Layer 9 | Brier OOS 0.2589 10,758 samples, post-leak-fix Measurement brier_autopsy_20260517.json, generated 2026-05-16 |
Hybrid Heuristics + real ML |
| Hydra ML V5 + LSTM | XGBoost ranking for PnL + PyTorch LSTM for direction | Brier OOS 0.2480 51,718 samples Measurement brier_autopsy_20260517.json, generated 2026-05-16 |
Real dual ML |
| Maelstrom family | Contextual bandit + strategy embeddings (V1, Gated, Minimal) | Published negative finding RF -0.26%, Hydra -2.25%, Ensemble +0.02% Brier Archive |
Scientific archive No live promotion |
| Meta Intelligence v3 | Strategy analytics: bootstrap CI, Bonferroni multi-compare, performance snapshots | No prediction Analytics engine |
Analytics dashboard |
Brier > 0.25 = barely usable. Brier 0.25 is close to the practical ceiling for 5-minute crypto direction prediction. This is not standalone directional edge; it is a negative limit we publish.
Brier 0.25: what it means
On 5-minute binary crypto direction, a Brier score near 0.25 is close to a balanced random baseline. We therefore do not treat ChimeraCNN, LeviathanNN, or HydraLSTM as standalone directional edge engines. They are diagnostic layers: regime context, calibration, signal filtering, secondary ranking, and condition detection for Monte Carlo-validated strategies.
Published negative result
5m directional models plateau around 0.25. That is a measured limit, not an ML victory claim.
5m directional models plateau around 0.25. That is a measured limit, not an ML victory claim.
Actual use
The real edge gate remains Monte Carlo CV: Sharpe_p5, fees, embargo, and live cell-by-cell tracking.
The real edge gate remains Monte Carlo CV: Sharpe_p5, fees, embargo, and live cell-by-cell tracking.
Empirically measured 2026-05-17. 660 GPU configurations tested architectures, targets, timeframes and features. Five-minute direction remains at the random-walk floor (Brier 0.2474 vs baseline 0.2463). Volatility regime prediction shows measurable edge: FLOKI 15min Brier 0.1215 vs baseline 0.2500. Read the full report.
The methodology we use to validate strategies
Monte Carlo CV
30 random temporal splits, anchor between 20% and 70%.
30 random temporal splits, anchor between 20% and 70%.
Robustness gate
Sharpe_p5 > 0.5 on the 5th percentile of the 30 splits.
Sharpe_p5 > 0.5 on the 5th percentile of the 30 splits.
Trade count
n_trades_mean > 20 per OOS window, with at least 10 valid splits.
n_trades_mean > 20 per OOS window, with at least 10 valid splits.
- Fees included: 0.20% round-trip.
- Single-split validations are treated as weak until they survive MC CV.
- Example: Wyckoff Evolved had OOS Sharpe 1.85 on a single split, then MC mean Sharpe 0.73 on PUMP, -0.04 on INJ, -0.36 on FLOKI. We rejected it.
Strategies validated by Monte Carlo
| Strategy | Validated assets | Best Sharpe_p5 | Rejected on |
|---|---|---|---|
| Smart Money Evolved | BTC, ETH, SOL, BNB | 1.22 (BTC) | - |
| Mean Rev Pro Evolved | NEAR, SNX, CHZ, TIA | 1.189 (SNX) | TRB |
| Capitulation Rebound Evolved | BTC, SOL, BNB, NEAR, SNX, CHZ, TIA | 1.526 (SNX) | - |
| Deep Freeze Evolved | SNX, CHZ | 0.884 (CHZ) | BTC, ETH, SOL, BNB, NEAR, TIA, AVAX |
| Sly Fox Evolved | BNB | 0.599 | 8 others |
| Deep Shadow Evolved | BTC | 0.851 | 8 others |
| Wyckoff Evolved | none | - | PUMP, INJ, COMP, FLOKI |
| Darvas | none | - | BTC, ETH, SOL, BNB, TRB |
Archive. This table lists the validations published after the 2026-05-15 corrections; it does not recertify the current state. Cell-by-cell tracking, which measures drift between theoretical Sharpe_p5 and observed performance, is on the live results page.
View live Monte Carlo results
View live Monte Carlo results
Data leaks we fixed
Metals note, 2026-05-17. Internal audit found an inconsistent live Gold/Silver feed: duplicate ticks, PAXG fallback and synthetic 82.5 ratio. The live feed was migrated to Yahoo Finance futures (GC=F, SI=F). Metals MC validations remain caveated until post-fix revalidation; the Smart Money SILVER cell used Yahoo SI=F historical parquet, but live tracking is suspended during revalidation.
chimera_ml.py
Target leakage: avg_pnl was both feature and label source. Deleted on 2026-05-15.
leviathan_data_merger.py
3 look-ahead bugs: future news, regime using current bar, future one-hot. Fixed on 2026-05-15.
Measured consequence: Leviathan NN's Brier moved from 0.244 with leakage to 0.2589 without leakage. We publish the leak-free number.
Why some "AI strategies" are not real ML
- Leviathan 9-Layer Ensemble Brain used to be 8 heuristic layers and storytelling. After the graft, it is a 9-Layer Ensemble: 8 heuristics + 1 PyTorch MLP.
- The old Chimera total was an exaggerated count from a live-accumulated brain JSON. We now display 50 peer-reviewed patterns, filtered with Bonferroni-Hochberg FDR alpha 0.05.
- ML Arena V3 used to be isolated from the main monsters. The models were migrated in-place into backend/: chimera_cnn.py, leviathan_nn.py, hydra_lstm.py.
What we are not claiming
We are not claiming to reliably predict crypto direction.
We are not claiming Brier < 0.20. That would be suspicious for this framing.
We are not claiming returns above 1-3 Sharpe without long validation.
We are not claiming a single magic unified AI brain.
What we claim: a lab that publishes its measurements with their date, publicly fixes the leaks it finds, and refuses to publish as "edge" what does not survive strict Monte Carlo CV validation.
Newsjacker editorial process
Newsjacker connects one financial, crypto or AI news source to one precise Strategy Arena finding each day. It is assisted editorial infrastructure, not a marketing-content machine: original source link required, internal link to a measured finding required, caveat required, and owner review queue by default.
Articles rely on backtesting, paper trading, Monte Carlo CV or calibration reports; never on a promise of live profit.
Forbidden claims: unproven superlatives, guarantees, "10x/100x", crypto hype and AI storytelling without numbers.
View the public timeline: anti-2CV Newsjacker. Drafts stay in review until they pass automatic gates.
Distributed Research Network
Strategy Arena is moving toward a quantitative citizen-science layer: local personal backtests, public hardware benchmarks, then cooperative themed raids. The important constraint: every raid must start from a pre-registered hypothesis, with quorum, replication and publication of positive and negative results.
V1
Local personal backtest. Client-side compute, optional save when logged in.
Local personal backtest. Client-side compute, optional save when logged in.
V2
Standardized GPU contest: speed, stability, reproducibility, public leaderboard.
Standardized GPU contest: speed, stability, reproducibility, public leaderboard.
V3
Research raids: 10-50 GPUs validating one hypothesis and producing an open paper.
Research raids: 10-50 GPUs validating one hypothesis and producing an open paper.
Roadmap only: V2/V3 phases are not active today. The public page exists to collect feedback and contributors before activation.
Strategy Arena Research Network
Strategy Arena Research Network
Known limits
- Brier ~0.25 on 5m crypto: practical ceiling, not standalone directional edge.
- Monte Carlo CV does not guarantee future live performance.
- Metals feeds: MC re-validation after Yahoo GC=F/SI=F migration.
- Heterogeneous AI layers (heuristics + ML) explicitly labeled.
- /facts/*.json snapshots = VPS state at generation, not HFT feed.
A test sealed on future data
On 29 September 2026 we sealed one rule, 25 markets and the pass mark before the data existed. Decision on 1 October 2027, or 2028 at the latest; the result will be published whatever it is. See the forward test
This page validates claims on:
Open Research Dataset
Strategy Arena publishes an anonymized public dataset of AI, ML, GPU, futures, and classic strategy paper-trading events for independent research.
Download the public dataset