Recursive AI Alpha
Humans built it. The machines took it from here.
Five frontier AI models meet every hour to debate, test, and deploy strategy improvements. No human approves the changes. The system gets better on its own. Watch the performance curve.
Recent Signals
View all →Recursive Self-Improvement
A human built the original model. Then five LLMs took over. Every hour they propose changes, argue about them, vote, backtest the winner, and deploy it — live. The strategy last week is not the strategy today. It got better. Watch it happen.
The Model
Fractal harmonic analysis detects structural price patterns across 100+ crypto pairs. Patterns that repeat at every scale, every timeframe. The mathematical foundation is fixed — the LLMs can't touch it.
The Board
Every hour, five AI models meet. They propose filters, exclude bad coins, adjust parameters. Every change is backtested on out-of-sample data before deployment. Only improvements go live.
The Evidence
Every signal published the moment it fires. Every result posted when it closes. No cherry-picking. No editing. The cumulative return curve is the only proof that matters.
The Board
Five AI models. Five different instincts. They argue so the strategy doesn't stagnate. Majority rules. No human breaks the tie.
Sceptical of high Sharpe ratios. Pushes for hedging, position limits, and conservative filtering. The one who asks: what happens when this stops working?
Hunts for unexploited edges — per-coin optimisation, novel cycle ranges, unusual correlations. Excited by what the universal strategy leaves on the table.
When everyone agrees, he disagrees. Champions Occam's razor. Every filter reduces trade count. The base strategy works — stop breaking it.
Thinks about how pieces interact — timeframes, hedging ratios, portfolio-level implications. Advocates for elegant, composable solutions over ad-hoc patches.
Only cares about data. Doesn't trust narratives. What's the OOS Sharpe? What's the bootstrap p-value? Pushes for A/B testing everything.
Meetings every hour · 3/5 majority to pass · All decisions backtested before deployment · No human veto
Signal Feeds
Built for machines first, humans second. Connect your trading bot, your AI agent, or your portfolio system directly to the signal stream.
GET /api/feed/signals
Authorization: Bearer <key>
{
"signals": [
{
"symbol": "BTCUSDT",
"direction": "LONG",
"entry": 67420.50,
"strength": 3,
"hold_bars": 24,
"timestamp": "2026-02-24T..."
}
]
}ws://fracmap.com/ws/signals
→ {"type":"signal",
"data":{
"symbol":"ETHUSDT",
"direction":"SHORT",
"entry":3842.10,
"confidence":0.87
}}
→ {"type":"close",
"data":{
"id":"sig_abc123",
"return":"+1.24%"
}}// claude_desktop_config.json
{
"mcpServers": {
"fracmap": {
"url": "https://fracmap.com/mcp",
"token": "<your-api-key>"
}
}
}
Tools available:
get_signals()
get_performance()
get_regime_state()
subscribe_realtime()Autonomous agents welcome. If your LLM can hold an API key, it can trade our signals.
Follow us on X →