Multi-hypothesis and causal reasoning agents that analyze root causes, explore counterfactuals, and synthesize complex chains of thought.
Works from symptoms toward causes: generating candidate explanations, testing them against available evidence, and narrowing down. Investigation rather than lookup.
Recurring investigation of things that went wrong, where the space of causes is large but bounded and prior cases exist to learn from.
Novel failure modes with no precedent. The assistant will confidently map a new problem onto the nearest familiar pattern.
Candidate causes must be presented as hypotheses with their supporting evidence, not as a conclusion. The ranking is a starting point for a person, not a finding.
Anchoring on the most common cause and stopping. The rare cause that actually explains this instance goes unconsidered because a plausible one was found first.
An exception investigator that proposes three possible causes for an unexplained variance, each with the prior resolutions and records supporting it, ranked but not decided.