One event.
Infinite ripples.
See them first.

Edward Lorenz's butterfly effect — the sensitivity of complex systems to initial conditions — is not metaphor in financial markets. It's mechanism. A factory fire in Taipei cascades into chip shortages in Austin, which delays GPU shipments to hyperscalers, which pushes cloud compute costs up, which shifts AI training budgets, which moves capital markets.

Traditional trading systems analyze events in isolation. ButterflyEffect.ai maps the full causal chain — tracing influence, direction, and confidence across every node in the ripple. The platform doesn't predict price; it predicts causation, then derives price direction from the causal graph.

Each chain consists of 8-12 nodes, where every node carries:

• Influence (0–1): How much market impact this node has
• Direction Score (-1 to +1): Bearish through bullish for specific assets
• Confidence (0–1): Model certainty in this causal link
• Asset Tags: Specific tickers affected (e.g., NVDA, TSM, MSFT)
• Time Horizon: Immediate (hours), short (days), medium (weeks), long (months)

The chain is the backbone of all decisions. Without a causal thesis, no trade is opened — regardless of what other signals say. This is enforced architecturally via REQUIRE_CHAIN_THESIS = True in the ensemble combiner.

SYNAPSE (Synaptic Network for Adaptive Prediction, Signals, and Execution) is a multi-layer intelligence system built for autonomous operation:

Tech Stack:

Inspired by Renaissance Technologies' approach of combining thousands of weak signals, SYNAPSE uses a 10-component weighted ensemble. Each component returns a directional score (-1 to +1) and a confidence (0 to 1). Components are confidence-weighted — a high-confidence signal counts more than a low-confidence one.

Adaptive Weights: The ensemble periodically recalibrates weights based on component hit rates from the last 50 trades. Components with higher accuracy get boosted (up to 2x), underperformers get dampened (down to 0.5x). Weights are re-normalized to sum to 1.0 after adjustment.

Every position is sized using the Kelly Criterion — the mathematically optimal bet size given a known edge:

Dynamic Exit Thresholds — derived from the chain's time horizon and influence:

Signal Decay: Exponential decay based on half-lives (immediate=4h, short=3d, medium=2wk, long=2mo). Old signals lose strength — the system doesn't chase stale theses.

Transaction Cost Awareness: Kyle's lambda market impact model estimates spread + impact + commission. Trades are rejected if expected alpha doesn't exceed 2x total costs (safety multiplier). This prevents overtrading on weak signals — a key lesson from Renaissance's obsession with execution costs.

SYNAPSE runs a continuous 3-tier learning loop:

Current Performance:

• Walk-forward accuracy: 57% (honest blind test — above random)
• Historical backtest: +77.7% return on 500+ events (2018–2025)
• Training data: 220+ curated examples with 70+ fine-tune cycles completed
• Local model: synapse-expert-14b (Qwen2.5 14B, Q4_K_M quantized, ~9GB)

Accuracy is transparently reported and continuously calibrated. The system knows what it's bad at and specifically trains to fix those weaknesses.

Renaissance's Medallion Fund famously used statistical arbitrage as a core strategy. SYNAPSE implements Engle-Granger cointegration testing across 15+ candidate pairs:

• Cointegration Test: p-value threshold 0.05 — only statistically validated pairs qualify
• Z-Score Entry: Enter when spread diverges by >2.0 standard deviations
• Z-Score Exit: Close at <0.5 standard deviations (mean reversion)
• Stop Loss: Close at >3.5 standard deviations (relationship breakdown)
• Half-Life: Computed per-pair — only trade pairs with half-life <60 days (reasonable reversion speed)

Pairs signals are fed directly into the auto-trader alongside ensemble signals, creating a diversified signal mix that doesn't depend solely on directional bets.

The tradeable universe spans 340+ instruments across 23 sectors, automatically refreshed from S&P 500 indices:

Instrument types include individual equities, sector ETFs, bond ETFs, commodity ETFs, index ETFs, and country ETFs. The universe is designed to capture the full ripple surface of any global event.

The edge isn't prediction — it's causation mapping.

Most quantitative systems model price directly. They look at what happened and extrapolate forward. ButterflyEffect.ai is fundamentally different: it models why things happen, then derives price direction from the causal structure.

This approach has several structural advantages:

• Explainability: Every signal traces back to a specific causal chain with named nodes and reasoning. This isn't a black box — you can interrogate every decision.
• Adaptability: New event types (pandemics, policy changes, supply shocks) don't break the system because they're analyzed through causal reasoning, not historical pattern matching.
• Anti-fragility: Regime changes (risk-on → risk-off) are detected by the HMM and the system adapts. Black swan events get their own causal chains.
• Small Capital Advantage: Unlike billion-dollar funds constrained by liquidity and market impact, a smaller account can enter and exit positions without moving the market — accessing the same alpha at lower cost.

Renaissance's Medallion Fund proved that combining many weak signals with rigorous execution and continuous learning can produce extraordinary returns. ButterflyEffect.ai applies this philosophy with a unique twist: the butterfly causal chain provides structured reasoning that purely statistical approaches lack.