rubric-forecasting

To generate structured forecasts with auditable probability calibration, provide a JSON input with your question, options, resolution rule, and evidence. The engine designs a topic-specific rubric and scores evidence deterministically into normalized probability outputs.

Rubric scoring for forecasting separates semantic judgment from deterministic computation, preventing LLMs from hand-calculating probabilities. It maps evidence to rubric dimensions and computes scores via a Python engine, yielding auditable numeric results with explicit evidence tracking.

To perform sensitivity analysis on forecasted probabilities, the engine computes sensitivity deltas during the scoring process. It evaluates evidence quality, applies dependency penalties, and aggregates scores to check the robustness of your normalized probability outputs.

Yes, you can use JSON input to track evidence for binary event forecasting. The input contract accepts structured evidence where each item maps to options, stance, strength, and rubric dimension scores, returning results with an evidence ledger.

The best way to normalize forecast scores across multiple options is using softmax normalization. The deterministic engine aggregates weighted dimension scores for each option and applies softmax to produce normalized, auditable probability distributions.

Your multi-option forecast requires a YAML-defined entrypoint to establish the JSON input contract and invoke the Python engine. This ensures deterministic computation of evidence-to-option attribution, weighted scoring, and sensitivity analysis without relying on model math.