One automation path is failing, duplicating work, routing incorrectly, or disappearing without a useful error. I turn that symptom into a bounded, evidence-backed repair decision. We define one path: one trigger, one essential decision or transformation, and one expected result. I review sanitized artifacts, write the acceptance test, inspect or reproduce the failure with synthetic cases, and return ranked findings plus the smallest credible repair options. My linked public work is a self-directed, synthetic, non-production demonstration, not paid client history. It includes inspectable n8n JSON, companion Node.js logic, passing CI, a browser-local reliability scanner, and a sample diagnostic showing how confirmed findings are separated from inferred risk. Larger QA and reliability-test packages are available after an inquiry when the supplied evidence supports a responsible scope.