xrtm 0.9.0

The xRtm Framework (Generative Forecasting).

XRTM v0.9.0 — AI for Event Forecasting

XRTM runs event-forecasting workflows from the command line. It uses a deterministic baseline by default (no API keys needed) and supports any OpenAI-compatible endpoint for real LLM forecasts.

Install

pip install xrtm

Python >=3.11,<3.14.

Quick Start

# Deterministic baseline (no API key)
xrtm start

# With a real LLM via any OpenAI-compatible endpoint
xrtm start --provider openai --model deepseek-v4-pro --base-url https://api.deepseek.com

# Inspect results
xrtm runs show --latest

Commands

Command	What it does
xrtm start	Run forecasts (deterministic or real LLM)
xrtm demo	Quick 2-question deterministic demo
xrtm doctor	Check Python, packages, imports
xrtm runs show --latest	Inspect the most recent run

Providers

Provider	API key needed
deterministic (default)	None
openai / openai-compatible	OPENAI_API_KEY

Set OPENAI_API_KEY, OPENAI_BASE_URL, and OPENAI_MODEL in your environment or .env file.

Run Artifacts

Each run produces 11 artifacts in runs/<run-id>/: run.json, eval.json, train.json, forecasts.jsonl, report.html, blueprint.json, provider.json, questions.jsonl, events.jsonl, graph_trace.jsonl, run_summary.json

XRTM Ecosystem

Package	Role
xrtm-data	Schemas & question sources (real-binary corpus, Polymarket, Metaculus)
xrtm-eval	Scoring (Brier, ECE, LogScore)
xrtm-forecast	Runtime engine (agents, providers, topologies)
xrtm-train	Backtesting & optimization
xrtm	Product CLI (this package)

License

Apache 2.0