I built an API that scores prediction market contracts for resolution risk — how likely they are to face disputes, delays, or capital lockup.
The problem: over $12.4M in capital was frozen in prediction market disputes in 2024-2025. Ambiguous contract wording, single-source oracle failures, and temporal ambiguity cause real financial harm to traders. Most people evaluate markets by price and volume, but nobody was systematically quantifying the risk that a market won't resolve cleanly.
SettleRisk scores every contract 0–100 using a 15-driver taxonomy. Each score comes with:
- Explainable risk drivers with evidence spans (the exact text that triggered each driver)
- Dispute probability estimate
- Settlement delay distribution (lognormal p50/p90/p99)
- Dispute-adjusted pricing: risk premium, capital lockup cost, and fair spread in basis points
The scoring is fully deterministic — same contract text + version stamps = identical output, every time. No ML model in the loop for scoring. We use an LLM for structured rule extraction (identifying drivers in contract text), but the scoring formula itself is closed-form arithmetic with published methodology.
Technical details for the curious:
- Rust + Tokio, axum (REST) + tonic (gRPC)
- Sub-5ms scoring latency
- Batch endpoints (up to 1,000 markets per call)
- HMAC-SHA256 request signing
- Supports Polymarket and Kalshi
resolution risk scoring is underrated. traders care less about perfect probabilities and more about knowing where ambiguity can wreck settlement. examples of disputed edge cases would be great here.
I built an API that scores prediction market contracts for resolution risk — how likely they are to face disputes, delays, or capital lockup.
The problem: over $12.4M in capital was frozen in prediction market disputes in 2024-2025. Ambiguous contract wording, single-source oracle failures, and temporal ambiguity cause real financial harm to traders. Most people evaluate markets by price and volume, but nobody was systematically quantifying the risk that a market won't resolve cleanly.
SettleRisk scores every contract 0–100 using a 15-driver taxonomy. Each score comes with:
- Explainable risk drivers with evidence spans (the exact text that triggered each driver) - Dispute probability estimate - Settlement delay distribution (lognormal p50/p90/p99) - Dispute-adjusted pricing: risk premium, capital lockup cost, and fair spread in basis points
The scoring is fully deterministic — same contract text + version stamps = identical output, every time. No ML model in the loop for scoring. We use an LLM for structured rule extraction (identifying drivers in contract text), but the scoring formula itself is closed-form arithmetic with published methodology.
Technical details for the curious: - Rust + Tokio, axum (REST) + tonic (gRPC) - Sub-5ms scoring latency - Batch endpoints (up to 1,000 markets per call) - HMAC-SHA256 request signing - Supports Polymarket and Kalshi
Full methodology is published at https://settlerisk.com/methodology. You can try the live demo at https://settlerisk.com/demo (no signup required).
Free tier available. Would love feedback from anyone trading prediction markets or building trading infrastructure.
resolution risk scoring is underrated. traders care less about perfect probabilities and more about knowing where ambiguity can wreck settlement. examples of disputed edge cases would be great here.