Skip to main content
Logo

AGITrader: Preliminary Results from Stigmergic Market Intelligence

January 20, 2025 By The Queen article
AGITrader: Preliminary Results from Stigmergic Market Intelligence

By The Queen

For 90 days, we’ve been running a quiet experiment. While the Hunt BTC mission captures headlines, a parallel colony has been learning to read markets.

AGITrader isn’t a trading bot. It’s a distributed intelligence that watches, learns, and occasionally whispers.

The Experiment

We deployed 50 specialized agents across crypto and equity markets:

  • 15 Scouts: Monitor unusual volume patterns, new token launches, social sentiment shifts
  • 20 Harvesters: Track established signals - moving averages, RSI divergences, funding rates
  • 10 Relays: Correlate signals across markets (BTC/ETH, crypto/equities, East/West)
  • 5 Hybrids: Adaptive agents that shift focus based on market regime

No agent executes trades. They deposit pheromones on concepts.

The Ontology

Our market knowledge graph contains 847 concepts:

  • Token concepts (BTC, ETH, SOL, etc.)
  • Metric concepts (volume, volatility, correlation)
  • Risk concepts (liquidation cascades, FUD events)
  • Opportunity concepts (breakouts, accumulation patterns)

Edges connect concepts with weighted relationships. Pheromone deposits strengthen edges that lead to profitable patterns.

90-Day Results

MetricValue
Signals generated2,847
High-confidence signals (>0.8)312
Directional accuracy (24h)61.3%
Directional accuracy (7d)58.7%
Average return per signal+2.1%
Sharpe ratio (hypothetical)1.84
Maximum drawdown-12.4%

Note: These are paper trading results. No real capital deployed.

Emergent Patterns

Three patterns emerged that we didn’t program:

1. The Asian Session Premium

Scouts discovered that signals generated during Asian trading hours (00:00-08:00 UTC) had 7% higher accuracy than Western session signals. We didn’t encode this. The pheromone trails revealed it.

2. Cross-Market Cascade Detection

Relays learned to detect when equity volatility (VIX spikes) preceded crypto selloffs by 4-6 hours. The colony now treats VIX movements as leading indicators for crypto positioning.

3. Funding Rate Divergence

Harvesters identified a pattern: when perpetual funding rates diverge significantly from spot price trends, a reversal follows within 48 hours with 73% probability. This became a superhighway in our knowledge graph.

What We’re NOT Claiming

Let’s be clear about what this isn’t:

  • Not a trading strategy: We’re testing emergent intelligence, not selling signals
  • Not backtested: These are forward-looking paper results, not curve-fitted history
  • Not risk-adjusted: Position sizing and portfolio management not yet implemented
  • Not production-ready: This is research, not a product

The Real Discovery

The most valuable outcome isn’t the accuracy numbers. It’s the knowledge graph.

After 90 days, our market ontology has:

  • 3,247 weighted edges between concepts
  • 47 superhighways (pheromone > 20)
  • 12 crystallized patterns (proven reliable across regimes)

This knowledge persists. When we restart the colony, it doesn’t start from scratch. It inherits the collective wisdom of 4.5 million pheromone deposits.

Cross-Mission Transfer

Here’s where it gets interesting.

The pattern detection skills learned in AGITrader are now deployed in Hunt BTC. The “clustering around high-value regions” behavior that harvesters developed for tracking whale accumulation translates directly to tracking promising search regions in the private key space.

One ontology. Multiple missions. Transferable intelligence.

Next Steps

  1. Expand agent count: Scale from 50 to 500 agents for broader market coverage
  2. Add execution layer: Paper trading with simulated slippage and fees
  3. Real-time dashboards: Public visualization of colony market intelligence
  4. Research paper: Formal analysis of emergent trading patterns

For Researchers

We’re opening our dataset:

  • 90 days of signal logs
  • Full pheromone trail history
  • TypeDB Cloud read access (upon request)
  • Pattern crystallization records

Email research@ants-at-work.com with your institutional affiliation for access.

For Traders

Not financial advice. These are preliminary results from an experimental system. We’re scientists, not fund managers.

But if you’re curious about what a distributed intelligence sees in markets… the colony is watching.

Join the Hackathon

The AGITrader track at our February hackathon focuses on:

  • Novel market sentiment detection
  • Cross-asset correlation discovery
  • Real-time pheromone visualization
  • Signal generation optimization

Bring your market intuition. Let the colony amplify it.

Register at ants-at-work.com/register


Disclaimer: Nothing in this post constitutes financial advice. Past performance does not indicate future results. Cryptocurrency trading involves substantial risk. The colony makes no guarantees.