abductive-oracle

Returns the single most parsimonious hypothesis explaining a set of observations, using one of three sub-oracles selected by data type. Never guesses — returns Hypothesis.nothing when no consistent explanation exists.

Use When

• Selecting the best hypothesis from a candidate set given evidence
• Performing inference-to-best-explanation (IBE) over structured or natural language observations
• Scoring hypotheses with MCMC sampling, LLM-based reasoning, or constraint propagation
• Classifying hypotheses into accepted (+1), suspended (0), or rejected (-1) via fixed thresholds

Workflow

1. Collect observations E and define hypothesis space H with priors
2. Select sub-oracle based on data type (see Oracle Selection Policy)
3. Run inference to find H* = argmax P(H|E) subject to consistency, explanatory power, and minimality
4. Classify result using fixed trit thresholds: posterior > 0.70 → +1, > 0.10 → 0, else → -1

Type Specification