Hydra ML: 3 Machine Learning Brains Analyzing 1,115 Trades in Real Time

The meta-learner that learns from the learners

In Strategy Arena, 58 strategies trade in parallel. Some win, others lose. Hydra watches all the strategies and learns: which strategies work in which context?

It's not a trader — it's a meta-learner. It doesn't predict Bitcoin's price. It predicts which strategy will win the next trade.

How Hydra works

Data collection

Every 10 minutes, when the 58 strategies trade, Hydra records: - Which strategy traded - Direction (buy/sell) - 11 technical features at the time of the trade - Outcome: won or lost - Market regime (BULL/BEAR/NEUTRAL/VOLATILE)

After weeks of accumulation: 1,115 trades analyzed, 75 strategies studied.

Model training

Every 6 hours (automatic cron), Hydra trains a LightGBM model on the accumulated data:

• 892 trades for training, 223 for testing
• 11 features per trade
• Training in 8.5 seconds
• Accuracy: 100% on the test set

The model learns to distinguish winning trades from losing ones based on the technical context and the strategy.

The 10 most important features

The dominant feature is is_buy (score 188): certain periods are massively favorable to buys or sells. Distance from the high (dist_high_20) comes second — when price is far from its peak, buying becomes risky.

Interesting fact: RSI, which is Invictus's #2 feature, is only #10 for Hydra. The two systems look at the market from very different angles and are therefore complementary.

Top strategies by ML reliability

Hydra calculates an ML score for each strategy — the average probability that its trades will be winners:

Mean reversion strategies dominate — in NEUTRAL regime (the most frequent), mean reversion trades are the most reliable.

Analysis by market regime

Hydra doesn't settle for a global score. It analyzes each strategy by regime:

NEUTRAL (1,046 trades — 94% of total)

• Best: Mean Reversion Gold (100% WR), Safe Haven (100% WR), Gold/Silver Ratio (100% WR)
• Worst: Golden Breath Gemini (0% WR), Central Bank Momentum (0% WR)
• Takeaway: in a calm market, mean reversion strategies crush everything

BULL (32 trades)

• Best: DebateForge AI (60% WR, +0.16%)
• Takeaway: in a bullish market, collaborative strategies like DebateForge (5 agents that vote and mutate) outperform

VOLATILE (24 trades)

• Very little data yet — Hydra needs more trades in volatile regime to be reliable

BEAR (13 trades)

• Insufficient data — BEAR regimes are rare in the current sample

Hydra in the ecosystem

Hydra is one of the 7 layers of Leviathan — the super-general. With 39.7% accuracy on Leviathan's votes, it's the least reliable layer for predicting market direction. But that's not its job: Hydra excels at predicting which strategy will win, not whether the market will go up or down.

→ Prompt Forge: Hydra's recommendations are injected into every AI prompt. When Claude or GPT has to decide, it knows that "Hydra recommends Mean Reversion in NEUTRAL regime (score 92%)".

→ Collaborative Arena: Hydra votes as one of the 9 brains, alongside Chimera and the 6 LLMs.

→ Invictus: Hydra identifies the reliable strategies, Invictus identifies the deadly contexts. Together, they form a complete shield.

See Hydra in action

The /hydra page displays in real time: - The ML score for each strategy - Recommendations by regime - Feature importance - Training history (50 sessions)

The Genie Pantheon can also answer your questions — ask "Which strategy does Hydra recommend right now?" and 6 oracles debate.

Hydra ML is part of the Strategy Arena ecosystem. 100% educational — not financial advice.