postgres-performancelisted

# PostgreSQL Performance Use a **tight evidence loop**: `baseline -> bottleneck and source -> falsifiable cause -> smallest intervention -> comparable delta` Choose only the unresolved links needed for the user's decision. Evidence that places the constraint outside PostgreSQL is a valid result. Without a baseline, return a bounded collection plan instead of settings. Optimization lowers end-to-end workload cost while preserving correctness. Account for writes, tail latency, replicas, and maintenance when they can change the decision. ## Evidence bar - A diagnosis states the decisive observations, then ranks the leading bottleneck and workload source against the strongest competing explanation. - A recommendation connects the smallest intervention to a falsifiable cause and names its cost, failure signal, and rollback. - A proven optimization has fresh comparable evidence for the target and protected invariants. When evidence stops earlier, report the exact missing link and next bounded probe. Keep correlation labelled as correlation until a causal probe supports it. ## Authority - For answer, review, diagnose, or plan requests, preserve database state while inspecting available code, plans, telemetry, logs, and bounded read-only probes. - For change, fix, or optimize requests, make in-scope local changes and run non-destructive validation. Production DDL/DML, configuration changes, restarts, session cancellation, maintenance, and load tests each require