Incident response
How Nebbos detects, communicates, mitigates, and postmortems a production incident. The shape is deliberate: the substrate’s audit chain is the record, not a chat log.
Severity classes
| Severity | Impact | Notification |
|---|---|---|
| Sev 1 | Production down — substrate unreachable, audit chain broken, mass data-access failure | Every affected operator paged; status page declares immediately |
| Sev 2 | Major impairment — one functional surface degraded; core reads/writes still working | Affected operators notified within 1 hour; status page within 30 min |
| Sev 3 | Minor impairment — degraded performance on a subset of surfaces; workarounds available | Status page notes within 2 hours |
| Sev 4 | Question / feature request — not an incident but tracked in the same queue for measurability | — |
Detection
Every substrate surface emits synthetic-probe metrics every 60 seconds from three geographic regions plus continuous audit-chain integrity checks. Alerts fire before an operator would notice; the on-call rotation is paged within two minutes of a Sev-1 or Sev-2 trigger.
Operator-reported incidents open at enterprise@nebbos.ai with subject line prefixed [incident]. The response-time commitment for these matches the SLA.
Communication
Every Sev-1 and Sev-2 incident gets:
- A public entry on the status board within five minutes of declaration
- Direct email to every affected enterprise operator within one hour, naming what is affected and what to expect next
- Updates every 30 minutes (Sev-1) or hourly (Sev-2) until mitigated. Each update names the change since the last update — no filler
- A resolution email within 24 hours of mitigation, naming the corrective actions
Postmortem
Every Sev-1 and Sev-2 incident publishes a postmortem within seven days of resolution.
- Format: what happened, what the impact was, why it happened (five whys), what has already changed, what will change with names and dates attached
- Style: blameless
- Where: publicly viewable at status.nebbos.ai
The postmortem is the artefact the substrate learns from. Every corrective action lands as an entry in the governance corpus and (where relevant) an ADR. Regulator-facing: postmortems are part of the EU AI Act · Annex IV post-market monitoring plan.
What the audit chain records
Incidents themselves land in the audit chain as distinct entries: declared, updated, mitigated, resolved, postmortem-published. An auditor can walk the chain end to end and reconstruct every incident’s lifecycle without contacting Nebbos.