Performance Metrics

Performance Metrics is the comprehensive observability layer that exposes service health data from every component of the CueCrux platform through Prometheus, providing operators with the quantitative visibility they need to maintain reliability, diagnose issues, and plan capacity.

The metrics cover several dimensions of platform health. Latency histograms track the response time distribution for every service endpoint, from Engine query processing through FactoryCrux ingestion to VaultCrux decryption operations. Rather than simple averages, which can hide problematic tail latencies, histograms reveal the full distribution of response times, making it easy to identify p50, p95, and p99 latencies and detect degradation that affects only a small percentage of requests.

Error rate tracking monitors the frequency and type of errors across all services. Errors are categorised by type, severity, and affected component, enabling operators to distinguish between transient network issues, systematic processing failures, and capacity-related problems. Error rate trends over time reveal whether the platform's reliability is improving or degrading and help identify the root causes of reliability issues.

Retrieval quality scores provide a unique dimension of platform health specific to CueCrux's knowledge retrieval mission. These scores track the relevance of retrieved evidence, the diversity of sources in evidence sets, and the confidence levels of synthesised answers. Declining quality scores might indicate corpus staleness, embedding degradation, or retrieval parameter misconfiguration, alerting operators to issues that would not be visible through traditional infrastructure metrics alone.

Budget consumption metrics track CRUX credit usage across tenants, assurance modes, and service types. Operators can see real-time spending rates, project budget exhaustion timelines, and identify tenants or workflows with unusual consumption patterns. These metrics support both operational management and business analytics, informing pricing decisions and capacity planning.

All metrics carry structured metadata labels that enable flexible slicing and dicing. Labels include tenant identifier, service name, endpoint, assurance mode, embedding lane, and knowledge domain, allowing operators to drill down from platform-wide aggregates to specific tenant-service-endpoint combinations. Grafana dashboards provide pre-configured views for common operational scenarios, with the flexibility to create custom dashboards for specialised investigations.

Alerting rules define thresholds for each metric class, triggering notifications through OpsCrux when values cross into concerning territory. Alert routing ensures that the right team is notified based on the type and severity of the issue.

For the CueCrux platform, Performance Metrics provides the quantitative foundation for operational excellence. It ensures that the platform's promise of reliable, transparent, evidence-backed answers is backed by rigorous operational monitoring.