Good Judgment vs
SimpleFunctions.

Good Judgment is a forecasting service descended from the academic Good Judgment Project. It produces probability estimates through trained Superforecaster panels — vetted human experts who reason through specific questions. SimpleFunctions is an API-first platform that aggregates live prediction market prices from Kalshi and Polymarket, computes indicators and calibration scores, and exposes a thesis system, autonomous trading agent, and 56-tool MCP server. Good Judgment is a human-panel service; SimpleFunctions is a machine-readable, code-first agent stack.

Based on publicly available information, Good Judgment does not publish a documented public REST API. Access to their forecasts is through the Good Judgment Open community platform or commissioned Superforecaster engagements — neither of which exposes a standard programmatic interface. SimpleFunctions exposes a fully public REST API, CLI, and MCP server with no auth required for read endpoints, documented at simplefunctions.dev.

POST a natural-language sentence to /api/thesis/create. SF decomposes it into a causal tree of testable sub-claims, maps each sub-claim to live Kalshi and Polymarket contracts, propagates probabilities up the tree, and launches an evaluation heartbeat: news scan, price refresh, milestone check, LLM eval, confidence update. You can inject signals at any time via /api/thesis/{id}/signal, and public theses are forkable. No other prediction market product exposes a system like this.

Autopilot is an autonomous trading agent powered by a 1M-context LLM drawing on 13 data sources — live prices, orderbook depth, computed indicators, thesis states, regime labels, and more. Before any execution it runs a 7-gate risk cascade: kill switch check, position limits, drawdown gate, regime check, and additional guards. It is designed for agents and quantitative researchers who want execution to be as structured and auditable as the research layer above it.

SF does not employ human forecasters. Its prices reflect live prediction market consensus on Kalshi and Polymarket. SF publishes its own Brier scores at /api/calibration — Kalshi 0.20, Polymarket 0.12 on T-24h price, past 90 days — so you can evaluate the accuracy baseline yourself with a curl call. Whether market consensus or trained human panels performs better depends on the question domain and time horizon; the empirical literature on this remains active.

Good Judgment wins when you need qualitative human reasoning on bespoke questions — geopolitical scenarios, low-base-rate events, or questions where no liquid prediction market has formed. Their Superforecaster methodology was designed precisely for that use case. SimpleFunctions has no substitute for a commissioned human panel on a question with no active market; the thesis system requires at least some tradeable sub-claims to propagate probabilities.

/api/calibration returns live Brier scores by venue, category, and price bucket, computed from the past 90 days of resolved contracts — Kalshi 0.20, Polymarket 0.12 on T-24h price. The endpoint is public and unauthenticated; you can re-verify with a single curl call. SF also releases daily datasets on HuggingFace and Kaggle under CC-BY-4.0. SF is the only prediction market API provider we are aware of that publishes a live, machine-readable calibration baseline.

Partially. SF indexes 48K+ active contracts on Kalshi and Polymarket, which cover a wide range of political, economic, and current-events questions. Good Judgment's Superforecaster panels can address bespoke, lower-liquidity questions for which no market exists. Where liquid markets do exist for a question, SF will have real-time prices and computed indicators; where no market exists, SF has nothing to offer and Good Judgment's panels become the relevant tool.