Cognitive observability
Observability focused on cognition as a distributed process. Covers memory retrieval decisions, arbitration margins, policy drift vectors, consolidation results, and self-modification proposals using OpenTelemetry under the cog.* namespace.
Traditional observability - metrics, logs, traces - is designed for distributed services where the question is 'why is this endpoint slow?' Cognitive observability extends this to answer questions like 'why did the system retrieve the wrong memories?' or 'which agent role is consistently producing low-confidence proposals?' The cog.* namespace in OpenTelemetry adds semantic conventions specific to cognitive events: cog.retrieval.latency, cog.arbitration.margin, cog.policy.drift_magnitude, cog.calibration.error, cog.working_memory.slots_used. These metrics turn the cognitive loop into something as instrumentable as a microservice. Arbitration margins tell you how close decisions were - a consistently narrow margin between the winning and second proposal indicates the system is operating near its decision boundary. Policy drift vectors show which behavioral dimensions are shifting most rapidly. Calibration error trends reveal whether agent confidence is becoming more or less accurate over time. These signals feed dashboards, alerting rules, and the meta-cognition engine's watchdog agents.