Pattern library
The OpenSearch-backed store of system-agnostic operational patterns. Each pattern defines a signature (co-occurring primitives), precursors (early-warning primitives), an outcome description, and ordered interventions. Confidence is updated by the reinforcement feedback worker based on observed recommendation outcomes.
The pattern library is the accumulated operational intelligence of the system expressed as reusable, transferable knowledge. Each pattern is a structured record with four key components. The signature is the set of operational primitives that, when detected co-occurring, indicate this pattern is active - for example, BACKPRESSURE_ACCUMULATION + QUEUE_GROWTH + TAIL_LATENCY_EXPANSION together identify a consumer-saturation pattern. The precursors are the early-warning primitives that tend to appear before the full signature manifests, enabling proactive intervention rather than reactive response. The outcome description is what happens if the pattern runs to completion without intervention. The interventions are an ordered list of recommended actions, from lightest (scale consumer replicas) to most disruptive (partition rebalance, producer throttling). Each pattern carries a confidence score updated by the reinforcement feedback worker using exponential moving average over observed recommendation outcomes: when a recommended intervention resolves the situation, confidence rises; when it does not, confidence falls. The transfer layer means this confidence history is not wasted when moving to a new infrastructure environment - the pattern library carries it forward, requiring only a new system mapping to activate.