Incident & Reliability joins Datadog monitoring, Bugsnag application evidence, incident command, governed Engineering fixes and recovery verification without collapsing their safety boundaries.
Unified workflow
A live reliability monitor correlates provider signals to incidents on a recurring loop (about every 60s); you can also sync active Datadog monitors or record a bounded manual signal. Cards show severity, service, environment, release, fleet distribution, spike count and affected users when the provider reports them.
Datadog and Bugsnag signals can join one open incident when service, environment, release, team and timing cross a conservative evidence threshold. Weak matches remain separate; accepted matches record their score and reasons.
Assign an owner and incident commander from the configured Jira project roster. Advance the governed lifecycle with evidence and follow the incident-response playbooks retrieved from the dedicated RAG collection.
Code-shaped evidence creates a read-only Engineering investigation. Host, Agent and configuration evidence records an Operations handoff. Neither action patches, deploys or resolves automatically.
A patch or PR is not resolution. Confirm Datadog health, Bugsnag error reduction and the configured stable recovery window before resolving, then complete the review and corrective actions.
FABRIC Intelligence
Forge retains normalized facts and authoritative links—not raw Datadog logs or complete Bugsnag payloads. Provider credentials remain server-side and provider data stays authoritative in its source.
Analysis retrieves diversified chunks only from the incident_response collection. The result separates observations from likely causes, missing evidence, verification steps, response recommendations and cited playbooks.
Fleet groups, request spikes and affected users are different units. Missing spike or user values remain “Not reported”; FABRIC must not convert alerting hosts into users or invent impact.
FABRIC analysis is advisory. Lifecycle transitions, ownership, mitigation, Engineering plans, patches, approval, publication and resolution remain explicit governed decisions.
Governed fixes
A forge.incident-handoff.v1 package carries up to eight bounded correlated signals, severity, service, environment, release and source references into Observe and Ground. Datadog and Bugsnag evidence enters as the internal trust tier (vetted systems), distinct from untrusted free-form text. The Engineering state machine remains independent — the handoff feeds the full 17-state governed engine (observed → grounded → planned → patching → patched → reviewing → reviewed → validating → validated → awaiting_approval → approved → committed → draft_pr_created, then release observation → verification), which stops at a draft PR.
Setup
Save one shared service-access token under Settings → Connectors. Dev, Product and Production reuse it; each workflow keeps its own enable flag, monitor tags, environments, Jira project and recovery window.
Register incident communication, escalation, response, checklist, LiveOps and on-call pages in the incident_response collection, ingest them, and confirm source freshness before relying on citations.
Shared language
SEV 1 through SEV 4 use one governed definition across teams. The incident detail panel includes a collapsible legend plus FABRIC recommendations for severity, impact, ownership and next action. Read the full Incident Severity Model →