anonicml2026/ANON_MODEL

Anonymous Outcome-Supervised Forecasting Model

Description

This is a decoder-only language model fine-tuned from Qwen3-32B for probabilistic forecasting of real-world events under uncertainty.

The model is trained using outcome-based supervision, where learning signals are derived from the eventual resolution of real-world events. Predictions are generated from causally masked inputs, and rewards are computed retrospectively using proper scoring rules once outcomes resolve. No human annotation or intermediate labels are used.

Intended Use

Research and experimentation involving:

• Binary event prediction
• Probabilistic forecasting
• Learning from delayed outcome-based feedback
• Calibration and uncertainty estimation

Limitations

As a fine-tuned derivative model, performance depends on the base model and training data. The model may produce incorrect or poorly calibrated predictions and should not be used in safety-critical settings without additional evaluation.

License

This model is a fine-tuned derivative of Qwen3-32B and is released under the Qwen3 License. All original license terms and conditions apply.