Unclassified

Observability

Observability in IntegrateWise is built around three behaviors: trace execution, measure health, and preserve accountability.

For canonical runtime and operations context, see:

  • docs/spec/INTEGRATEWISE_CANONICAL_SPEC_v1.0.md
  • docs/ops/RUNTIME_TOPOLOGY.md
  • docs/internal/operations/DEPLOYMENT_RUNBOOK.md

Correlation and tracing

Every execution thread should carry a unique correlation identifier from the edge boundary. This identifier should be injected into internal messages, async event steps, and adapter payloads.

Integration partners should:

  • preserve correlation identifiers in outbound payloads where supported
  • log them alongside handler and transport metadata
  • include them in outcome and webhook payloads when available

Provider health

Provider health states are exposed through registry telemetry. Health is evaluated along:

  • latency
  • error rate
  • saturation

Rolling windows and threshold rules are tenant-configurable. When health degrades:

  • registry shifts provider state to degraded
  • breaker logic may be tripped
  • Workbench displays explicit warnings
  • capability availability changes are reflected in the Capability Fabric surface

Execution telemetry

Hermes reports:

  • queued state
  • executing state
  • provider confirmation
  • timeout events
  • failure classifications
  • outcome events

Partners should not infer success from absence of failure. Always check provider confirmation and promotion outcome before treating an action as canonical.

Audit expectations

Every consequential action should emit audit events that preserve:

  • actor
  • authority scope
  • policy validation result
  • affected context
  • exact provider operations
  • execution-plan version
  • approval record
  • provider outcome
  • synchronization result

Audit logs are not debugging logs. They are accountability records.

Alerting and incident signals

Operators should monitor:

  • provider error rates and breaker state changes
  • governance denial or approval latency
  • approved proposal execution failure rates
  • sync conflict volume
  • sync lag beyond configured policy windows
  • cache staleness for critical projections

Toy alerts increase incident noise. Target alerts that indicate material risk, user impact, time sensitivity, or system integrity issues.

Partner incident support

When a partner integration causes incidents:

  • provide correlation IDs, approval tokens, proposal IDs, and execution-plan versions
  • preserve raw payloads and adapter responses within retention windows
  • do not alter logs after incident time without timestamped, documented, governed changes
  • notify the customer-facing support path through established channels