bayes-reasoner

Bayesian hypothesis balancing identifies the most probable root cause by updating beliefs based on evidence, reducing cognitive bias during incident investigations with sparse data. You initialize competing hypotheses, define diagnostic tests, and update probabilities as evidence arrives.

Probabilistic reasoning handles unlisted causes through automatic normalization and an explicit "Other" hypothesis option, ensuring unaccounted root causes maintain probability mass. This prevents overconfidence in listed hypotheses when conducting incident investigations with incomplete information.

Information-gain-driven recommendations identify the most informative next diagnostic step by calculating which test most effectively differentiates between remaining hypotheses. The system evaluates defined test likelihoods against current hypothesis probabilities to recommend the highest-value diagnostic action.

Yes, you can split complex hypotheses into subcategories during an active investigation using the split capability. This allows refining broad root cause hypotheses into more specific sub-hypotheses as evidence warrants, maintaining normalized probabilities across the expanded diagnostic state.

Inject new hypotheses when emerging evidence suggests a root cause not covered by existing options. The inject capability adds new hypotheses to the active state without resetting prior diagnostic progress, allowing the investigation to adapt as information expands beyond initial assumptions.

No external dependencies or cloud services are required. The probabilistic reasoning engine operates entirely locally via a Python CLI and stores investigation state in a local JSON file, making it suitable for offline incident response and sensitive diagnostic workflows.