Field Estimate
MethodologySample AnalysisGitHub
Open Source · MIT License

Geopolitical Forecasting
Research Platform

Field Estimate is an open-source tool for calibrated probability estimation on geopolitical and macroeconomic questions. It applies Le (2026) logistic recalibration — derived from 292 million historical trades across 9 domains — to correct for systematic over- and underconfidence in crowd probability estimates.

Analysts work through a five-step structured forecasting workflow: causal factors, reference classes, inside-view adjustments, community consensus comparison, and a structured decision log — producing a calibrated probability estimate with a documented analytical trail.

View on GitHubCalibration MethodologySample Analysis
⚗️

Calibrated Probability Estimation

Applies domain × time-horizon calibration slopes from Le (2026) — estimated from 292M trades across politics, economics, crypto, and six other domains — to correct systematic biases in crowd probability estimates.

See the methodology →
🗂️

Structured Analytical Framework

A five-step forecasting workflow — causal factors, reference classes, inside-view adjustments, community consensus, and a decision log — adapted from the structured analytic techniques used in professional intelligence analysis.

See a sample analysis →
🌐

Multilingual OSINT Synthesis

Aggregates open-source intelligence across English, Arabic, Farsi, and Hebrew sources. Claude synthesizes coverage into a structured brief with per-country and per-topic tagging, then matches relevant content to open questions.

View source →

What's in the codebase

Le (2026) recalibration
src/lib/le-calibration.ts
Brier score tracking
src/lib/calibration-stats.ts
Polymarket + Kalshi clients
src/lib/polymarket.ts, kalshi.ts
Claude forecasting pipeline
src/lib/anthropic.ts (~2400 lines)
Metaculus + Manifold aggregation
src/lib/community-forecasts.ts
FRED · BLS · BEA · EIA integration
src/lib/fred.ts, econ-releases.ts, eia.ts
pgvector semantic matching
96 Supabase migrations
Multilingual Tavily search
en / ar / fa / he