Economic Crisis Benchmark

Prime Intellect environment for predicting historical economic crises from anonymized macro panels.

Contents

• 80 examples by default: 40 crisis windows and 40 negative windows
• public data only, no API keys
• anonymized prompt with relative time axis only
• robust JSON-based scoring

Data sources

• JST Macrohistory Database R6: https://www.macrohistory.net/database/
• Maddison Project Database 2023: https://www.rug.nl/ggdc/historicaldevelopment/maddison/releases/maddison-project-database-2023?lang=en
• World Bank API helper included for optional extension, no credentials required

Run locally

uv sync --extra dev
uv run python -m economic_crisis_benchmark.build_dataset
uv run pytest

Prime Intellect

prime eval run economic-crisis-benchmark
prime env push economic-crisis-benchmark --path . --auto-bump

The environment module is economic_crisis_benchmark/economic_crisis_benchmark.py and exposes load_environment().