Signal Generation

Our signal engine works from a focused set of real, named data sources — live OKX and BingX candles, Coinglass derivatives data, CoinMarketCap global metrics, Whale Alert transfers, CryptoQuant exchange flows and curated crypto news — reviewed by a Claude AI layer for market bias and risk.

Each candidate passes through 8 independent checks. Any one of them can weaken the score or veto the trade outright.

• 1

  Trend Structure

  EMA 9/21/50 alignment, slope and crosses on 1-hour candles — the backbone of every direction call.

• 2

  Momentum

  MACD line/signal cross and histogram direction confirm the move has force behind it; RSI(14) flags exhaustion at the extremes.

• 3

  Volatility Assessment

  ATR(14) and Bollinger position size the exits: quiet coins get tighter TP/SL, wild coins get room to breathe.

• 4

  Volume Confirmation

  Current volume against its recent average — a real move needs participation, and thin breakouts score lower. Coins under $10M daily volume never enter at all.

• 5

  Market Risk

  Liquidation cascades, funding extremes and volatility spikes roll into a risk verdict that can cut leverage or halt entries.

• 6

  BTC Market Context

  Altcoin candidates are checked against what Bitcoin is doing — fighting the tide costs points, and sentiment extremes add caution.

• 7

  Market Structure

  Swing-point pivots map recent highs and lows, anchoring pullback entries near meaningful levels instead of mid-air.

• 8

  News & Red Flags

  An AI layer reads fresh crypto headlines per coin — hacks, delistings, regulatory hits — and puts flagged coins on a temporary skip list.

The checks combine into a 0–100 strength score plus a confidence value. Confidence below 50 skips the trade; strength decides how aggressive the position is allowed to be.

Take-profit and stop-loss are a 50/50 blend of ATR-derived distances and your configured percentages, bounded (TP 2–10%, SL 1.5–5%), with a minimum 2:1 reward-to-risk enforced. Both orders are placed on the exchange at entry.

• Transparency over black-box models
• Composable layers that can be individually tuned
• Fail-safe defaults when data is unavailable
• Human-readable confidence explanations
• We would rather miss a trade than force a bad one