Foresight

Foresight traded Polymarket crypto prediction markets using a dual-path signal architecture: reactive snipe windows for markets approaching close, and predictive early entry for markets where the signal fired before the crowd caught on. It ran on real capital. It is no longer running — the position was wound down and the wallet withdrawn. What follows is the architecture, not a live system.

The PolyEdge signal stack anchored every trade to InDecision's directional bias, then layered momentum confirmation across multiple timeframes. Each execution window had a strict TTL gate — 5m snipe (20s), 15m snipe (45s), 1h snipe (120s), daily (300s). Miss the window and the signal was logged but no trade fired. That prevented chasing: the discipline was structural, not willpower.

StateCheckpointer snapshotted regime, risk state, and Elliott Wave context to SQLite every 30 seconds. Sub-30-second restart preserved 100% signal quality — no cold starts, no lost state, no degraded performance after a crash. That was PR #114's core contribution: turning a fragile reactive system into a resilient predictive one.

Conservative mode ran by default (--mode conservative): it required MODERATE or above (score ≥ 70). The 🔥 STRONG SIGNAL log line did not mean a trade fired — the time-to-close check happened after scoring. Only signals that cleared every gate reached the Polymarket CLOB.

The system carries 1,970+ tests. The interesting engineering was never the returns — it was building something that could update its own strategy while holding live positions, and stop itself when the gates said stop.