Jev vs incident runbooks
An incident runbook is a sequenced response: page, mitigate, communicate. Jev can help pick which runbook from a page of alert text. It does not page people or roll back deploys.
Unofficial operate note. Official: code owns side effects. docs.typesafe.ai.
Comparison scope
PagerDuty/Opsgenie stay. Prometheus rules stay. Jev is optional on messy customer-reported incidents or multi-alert narratives.
Criteria that decide the architecture
| Axis | Jev (System One) | Incident runbooks |
|---|---|---|
| Page on-call | Your code after a gate | Runbook / paging tool |
| SEV label | Choice + confidence abort | Human IC |
| Mitigation steps | No | Wins |
| Numeric error rates | Code / metrics | Code |
Decision quality and control
Paging on a 0.55 Noul is how you burn the team. Fail to “page human IC” on timeout. Do not ask Jev to compare burn rates (math jaggedness).
Integration trade-offs
Alert text + recent deploy titles (filtered) → Choice over runbook ids + unknown → execute only on high confidence. Log everything in the incident timeline.
TypeSafe’s public models page lists jev-1.13 at $0.042 per million input tokens with output tokens free — a vendor claim, not a jev.pro measurement. Confirm on the models page before you bid.
When each approach fits
Prefer Jev when
- Customer-written incident text that does not match a metric name
Prefer Incident runbooks when
- The alert already maps 1:1 to a runbook
- Mitigation is the hard problem
Metrics page; Jev suggests a runbook; humans and automation execute the steps.
Name the runbook, do not run it
A runbook is sequenced response: page, mitigate, communicate. Jev does not page people or roll back deploys. Sketch: messy customer-reported text plus filtered recent deploy titles → Choice over runbook ids plus unknown on pinned jev-1.13.0 → execute only on high confidence → log the Choice and response.model on the incident timeline. Prometheus rules and PagerDuty stay. Numeric error rates stay in code.
The $0.042 / M input figure TypeSafe publishes is a vendor claim. Paging on a 0.55 Noul is how you burn the team — that cost is not on the models page. Jaggedness: do not ask Jev to compare burn rates or error-rate floats. Fail to “page human IC” on timeout. Jev does not generate the status-page paragraph.
How SEV classification pages the wrong people
- Low-confidence auto-page. Budget a floor. Below it, the IC owns severity. Score levels are rubrics, not “SEV-1 means 5.”
- Mitigation in the model. Steps stay in the runbook. A valid Choice is not “disk was full and we cleaned it.”
- Unfiltered deploy dumps. State rot plus possible secrets. Title list, not the changelog blob.
- 1:1 metric alerts through Jev. If the alert already maps to a runbook, skip the hop. You add latency and a 429 surface for no language problem.
- Schema-safe
unknownignored.unknownmeans run the human path, not “pick the first id.”
When not to use Jev here
Do not use Jev when the alert already maps 1:1 to a runbook, when mitigation is the hard problem, or when the only input is a numeric burn. Prefer runbooks + paging tools. Use Jev when customer-written incident text does not match a metric name.
What this page does not claim
-
Not an IR platform.
-
No MTTR claims.
-
Schema-safe output is not the same as factual correctness.
-
No independent bake-off numbers live here.
FAQ
Can Jev roll back Kubernetes? No. Your controller does, if policy allows.
Should SEV be a Score? Only if levels are descriptive rubrics, not “SEV-1 means 5.” Humans still own severity.
Can Jev post to Slack for me? Your code can, after a gate you accepted. The model returns a label, not a message.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Hub: Comparisons. Siblings: vs security automation, observability guide, safe defaults. Canonical: https://docs.typesafe.ai.