Obsidian is a crypto futures strategy prototyped in early 2025. Its strategy logic, decision flow, and trade management are structured around a tightly controlled framework designed to balance adaptability and robustness.

• Operates on intraday timeframes ranging from 15 minutes to 2 hours
• Entry signals are generated by an engine combining a small set of technical indicators with a reinforcement learning (RL) layer, trained directly on over 5 years of historical backtests
• Large Language Models (LLMs) — including OpenAI, Anthropic, and DeepSeek — are used to interpret short-term market context (typically the next few hours or end-of-day window) and dynamically determine Take Profit, Stop Loss, and Trailing Stop levels in real time

1) Narrow signal framework (indicator-light)

• Oscillator-based triggers designed to detect short-term overextension and mean-reversion opportunities
• Trend and momentum confirmation layers used to avoid structurally weak or low-quality market regimes

2) Reinforcement Learning layer (narrow ML, focused role)

• Accepts or rejects potential trade setups based on detected market regime
• Filters out choppy, low-signal, or unfavorable market conditions
• Dynamically reduces aggressiveness during periods of elevated volatility or unstable price structure
• Optionally modulates exposure sizing based on internal confidence metrics

3) LLM-managed exits (TP / SL per trade)

• Adaptive Take Profit and Stop Loss levels derived from volatility regimes and key market structure zones
• Context-aware calibration based on alignment (or divergence) between macro and short-term conditions
• Trade-by-trade risk management enforced within strict, predefined guardrails

4) Multi-timeframe blend (short-term + macro context)

• Short-term execution layer focused on precise timing and entry efficiency
• Daily macro trend context used for directional bias, filtering, and regime alignment