Title: Observability Strategy Version: 1.0 Owner: [TENANT_CONFIGURATION_REQUIRED — Platform/DevOps] Status: Draft Last reviewed: 2026-09-07 Next review: [TENANT_CONFIGURATION_REQUIRED] Reviewers: Architecture, Security, DevOps
Operationalizes ADR-005: OpenTelemetry-based traces, metrics, and structured logs across all services.
| Signal | Source | Backend (configurable) |
|---|---|---|
| Traces | HR Core API, Workflow Engine, Agent Orchestrator, Integration Adapters, MCP servers | OTel Collector → [TENANT_CONFIGURATION_REQUIRED backend] |
| Metrics | All services (RED/USE metrics) + AI-specific metrics | OTel Collector → Prometheus/Azure Monitor/Datadog |
| Logs | All services, structured JSON, redacted before export | Centralized log store |
correlationId, tenantId, subjectType/subjectId (where applicable), traceId/spanId — propagated across every service and MCP boundary per api-standards.md.
modelVersion, promptVersion, skillName, retrievalCount, documentVersion (RAG), tokenUsage (input/output), latencyMs, toolCallsMade, guardrailOutcome, approvalState, errorCategory. See ADR-005 for the redaction rule: never the raw prompt/response content or PII.
See logging-and-redaction-standard.md for the mandatory redaction pipeline applied before any signal leaves the service boundary.
Trace sampling ratio is configurable (TRACES_SAMPLING_RATIO in .env.example, default 0.1 in non-prod-critical paths); sensitive-action traces (approvals, employee conversion) are always sampled at 100% regardless of the global ratio.
See dashboard-specification.md and metrics-slos-and-alerts.md.
| Version | Date | Author | Change |
|---|---|---|---|
| 1.0 | 2026-09-07 | Documentation package generation | Initial creation |