I got tired of crypto "signal" tools that either (a) backtest with lookahead bias, or (b) never show you what actually happened after a signal fired. So I built one that does the boring, honest version: Pipeline: WebSocket ingestion (100 topics, 50 coins) → dual-horizon volume baseline (20/50-bar median, computed only on completed bars to avoid lookahead) → 4-EMA alignment + Wilder RMA RSI scoring (0-100) → a single Final Gate that hard/soft-blocks shorting parabolic movers and longing free-falling ones → forward outcome tracking at +15m/30m/1h/4h with real fees (0.11% round-trip), slippage (0.04%), and funding rate deducted. To avoid the classic "same signal re-fires every 5 minutes and inflates your sample" problem, repeated signals within a 30-min window get collapsed into a single event_id, and stats are reported at the event level (not signal level), with Wilson 95% confidence intervals and explicit sample-size tiers. It's currently frozen at baseline v1.0 — no logic changes for 1-2 weeks — so I can honestly evaluate: does the score correlate monotonically with win rate? Does the top-gainer short guard actually save more than it costs in missed upside? Is net expectancy positive after costs? Does it hold up across bull/bear/choppy regimes? Genuinely interested in HN picking apart the methodology — especially the event-clustering approach and the guard-rail logic. Live: https://www.marketradar50.com Stats/history: https://www.marketradar50.com/hi...

Radar 50

Real-time signal radar for 50 crypto markets — with proof

Real-time signal radar for 50 crypto markets — with proof