reasoning-abductivelisted

# Abductive Reasoning Generate best explanations from observations. The logic of diagnosis and inference to cause. ## Type Signature ``` Abductive : Observation → Hypotheses[] → Evidence → BestExplanation Where: Observation : RawData × Surprise → AnomalyDescription Hypotheses : AnomalyDescription → [PossibleCause] Evidence : [PossibleCause] × AvailableData → [ScoredHypothesis] BestExplanation : [ScoredHypothesis] → (Cause × Confidence × NextSteps) ``` ## When to Use **Use abductive when:** - Anomaly or surprise present (metric deviation, unexpected behavior) - Need to explain an observation (why did X happen?) - Incomplete data, must infer cause - Diagnostic context (errors, issues, failures) - Multiple possible explanations exist **Don't use when:** - Cause is already known (use causal execution instead) - Need to compare decision alternatives - Resolving stakeholder disagreements ## Four-Stage Process ### Stage 1: Observation **Purpose:** Transform raw data into structured anomaly description. **Input:** ```yaml observation: raw_data: "Conversion dropped from 12% to 7% in Q4" context: timeframe: "Q4 2025" baseline: "12% historical average" current: "7% observed" surprise_level: 0.8 # How unexpected is this? ``` **Process:** 1. Quantify the deviation (% change, sigma from mean) 2. Identify temporal boundaries (when did it start?) 3. Scope the phenomenon (which segments affected?) 4. Rate surprise level (expected vs unexp