🤖 AI suggested severity
A recommendation with evidence and limitations, never an automatic lifecycle transition.
Operations · Incident governance
Severity means the same thing.
Across every team.
Forge classifies incidents into four severity levels to standardize response, prioritization and escalation. FABRIC can recommend; accountable people review and decide.
1
Reference
Incident severity model
| Severity | Description | Business impact | Examples | Action |
|---|---|---|---|---|
| 🔴 SEV 1 · Critical | Complete production outage or critical security issue | Very high | Service unavailable, failed deployment, data corruption | Respond immediately, assign an incident commander, notify stakeholders, roll back or hotfix. |
| 🟠 SEV 2 · High | Major functionality unavailable or significantly degraded | High | Purchase failures, matchmaking failures, critical gameplay issues | Prioritize immediately, engage owners, and prepare a hotfix. |
| 🟡 SEV 3 · Medium | Partial degradation with an available workaround | Medium | Performance degradation, isolated gameplay bugs, UI issues | Investigate and schedule a fix in the current sprint. |
| 🔵 SEV 4 · Low | Minor defects with minimal player or business impact | Low | Cosmetic issues, text errors, minor UX improvements | Add to the backlog and address during routine maintenance. |
2
Evidence-based recommendation
How severity is assigned
FABRIC may recommend severity using bounded evidence about player impact, business and revenue impact, security risk, service availability, release status, stability and the reported number of affected users. Missing evidence remains missing—it is not converted into a score.
📊 Impact score
Separate 0–100 dimensions for Players, Revenue, Stability and Security. An absent value is shown as “—”, not zero.
👥 Suggested owner
A likely accountable team or person when the evidence supports it. Assignment remains a human action.
⚡ Recommended next action
The next evidence-backed response step, grounded in incident-management RAG and source telemetry.
Human review and override are required.FABRIC recommendations are hypotheses. The incident owner or commander confirms severity against authoritative telemetry and records any override in the timeline.
3
Governed lifecycle
Response workflow
Incident detected
Normalize bounded Datadog, Bugsnag or manual evidence.
AI recommends severity
Score only supported impact dimensions and state limitations.
Human review or override
Confirm severity from authoritative evidence.
Assign owner
Name the accountable owner and, for critical response, the incident commander.
Execute response
Follow the relevant playbook and governed Engineering or Operations route.
Validate resolution
Observe provider health through the stable recovery window.
Close and learn
Record cause, recovery evidence and corrective actions.