Structured forecasting tooling and Cursor Agent skill: the model interprets the question and organizes evidence; a Python engine performs weighting, normalization, and sensitivity analysis, and emits auditable JSON plus natural-language reasoning.

See INTRODUCTION.md for full concepts and workflow.

• Python 3.10+ (stdlib only at runtime; no third-party deps for execution)
• Tests and local dev: pip install -e ".[dev]" (below)

From the repo root (ensure the rubric_forecast package is on the path):

python scripts/rubric_forecast.py --input examples/polymarket_hormuz_geopolitics_input.json

Or as a module (also works after install):

python -m rubric_forecast --input examples/polymarket_hormuz_geopolitics_input.json

Read from stdin:

cat examples/polymarket_hormuz_geopolitics_input.json | python -m rubric_forecast

After pip install ., you can use the console entry:

pip install .
rubric-forecast --input path/to/input.json

A small stdlib-only server and static page call run_forecast over HTTP for quick manual checks:

python examples/web_demo/server.py

Open http://127.0.0.1:8765/ — load sample input, edit JSON, then run. Bind address is loopback only; do not expose this process to untrusted networks.

Add SKILL.md to Cursor Agent Skills (or your team’s .cursor layout). The skill states: do not hand-compute scores in chat; numeric results must come from this repo’s engine, and output must include an engine field.

Concepts and I/O details: INTRODUCTION.md.

Scripts under scripts/ such as evo_pipeline.py and evo_submit.py read a JSON config. Copy the example and adjust paths and secrets:

cp config/evo_config.example.json config/evo_config.json

Point wallets_file at your wallet JSON. Paths in the example are relative to the repository root; run those scripts from the repo root (or use absolute paths). The file config/evo_config.json is gitignored so machine-specific values are not committed.

pip install -e ".[dev]"
pytest

MIT