Thesis-Driven Prediction Market Trading: Why Causal Models Beat Signal Chasing

## The Signal-Chasing Trap Most prediction market bots work like this: watch for a price move, check if it crosses a threshold, buy or sell. The problem is obvious — you're reacting to effects, not causes. When oil prices spike, a signal bot sees "KXWTIMAX-T135 moved +8¢" and might buy. But *why* did it move? If it's a temporary supply glitch, the price will revert. If Iran just confirmed Hormuz mining operations, it's the beginning of a sustained move. The signal bot can't tell the difference. A thesis-driven agent can. ## What Thesis-Driven Trading Looks Like You start with a structured belief: sf create "Iran war escalation keeps oil above through 2026. Hormuz stays blocked, supply disruption persists, recession risk rises." The engine decomposes this into a causal tree: n1: US-Iran war occurs and persists 0.95 n1.1: Military conflict initiates 0.99 n1.2: Diplomatic resolution fails 0.96 n1.3: No rapid regime collapse 0.88 n2: Hormuz closure sustained 0.92 n2.1: Mines deployed successfully 0.99 n2.2: Mine clearing takes >3 months 0.99 n3: Oil price stays elevated 0.89 n3.1: SPR insufficient to offset 0.96 n3.2: Pipeline alternatives can't compensate 0.88 Each node has a probability. Each probability implies a price for related contracts. When market prices diverge from these implied prices — that's an edge. ## Why Causal Models Work Better **1. You know *why* you're buying.** "KXRECSSNBER-26 at 35¢, thesis implies 72%, edge +37¢" is a trade with a reason. If node n3 (oil stays elevated) weakens, you know to re-evaluate. A signal bot just sees +37¢ and has no framework for when to exit. **2. You can stress-test before committing capital.** # What if a ceasefire happens? sf whatif --set "n1=0.1" --set "n2=0.2" # Shows: confidence drops to 34%, oil edges reverse, recession edge halves Zero cost, instant answer. Try doing that with a signal-based system. **3. News becomes signal, not noise.** When WSJ reports "oil tankers pass Hormuz safely," the thesis engine checks which causal node this affects (n2.2 — mine clearing), recalculates downstream probabilities, and tells you the impact: "confidence -3%, recession edge narrowed by 8¢." A signal bot would see the headline and have no idea what to do with it. ## The Practical Workflow # Morning: check overnight changes sf dashboard # The heartbeat already scanned news, refreshed prices, re-evaluated # Investigate sf agent > what caused the confidence increase? > show me the gas edge with orderbook depth # Act sf buy KXAAAGASM-26DEC31-3.90 200 --price 8 The thesis does the thinking. You do the deciding. ## Getting Started npm install -g @spfunctions/cli sf setup sf create "your thesis" sf edges # see where markets disagree with you Documentation: [simplefunctions.dev/docs](https://simplefunctions.dev/docs)