AI Sentiment Shield

Before every BUY order, the engine checks the market's sentiment using two AI layers:

Layer 1: LunarCrush Social Intelligence We pull Galaxy Score and AltRank for the target symbol. These aggregate social volume, engagement, and sentiment across Twitter, Reddit, YouTube, and news.

Layer 2: LLM Analysis (Gemini Flash) The LLM receives the LunarCrush data + recent price action and produces a sentiment classification: BULLISH, NEUTRAL, or BEARISH with a confidence score (0-100).

Gate logic: - If sentiment = BEARISH and confidence > 70 → BUY is blocked - If sentiment = BEARISH and confidence 50-70 → position size reduced by 50% - Otherwise → normal execution

This prevents the bot from buying into panic or capitulation events where even technically valid signals tend to fail.

A secondary sentiment layer runs in shadow mode — it doesn't block trades but logs its decisions for validation:

• Aggregates headlines from CoinDesk, The Block, CoinTelegraph via our news-ingest pipeline
• Runs LLM classification on aggregated news context
• Compares its signal against actual 24h forward returns
• Once validated (4 consecutive weeks of positive alpha), can be promoted to live gating

This shadow-first approach ensures new AI features prove themselves before affecting real capital.

In backtesting over 2024-2025 data: - The sentiment shield would have blocked 73% of entries that resulted in >5% adverse moves within 24h - False positive rate (blocking profitable entries): ~12% - Net effect: +4.2% annualized return improvement, -2.1% max drawdown reduction

The shield is conservative by design — it's better to miss a few entries than to catch a falling knife.