Jev for operations
Operations teams win with Jev when exception text is the bottleneck. They lose when they let a Noul press an irreversible button.
Independent operate-pillar page — not TypeSafe’s runbook and not a WFM vendor. Confirm API behavior on docs.typesafe.ai. We do not issue keys. Replicas named here are not Jev and not endorsed.
Ops contract
- Action enum includes
review. - Timeouts → documented default (usually review or page).
- Reviewer UI shows questions + probabilities, not a fake rationale.
- Weekly sample of autos.
Where Jev sits
In front of queues (vs ticket queues), not instead of WFM/SLA tools.
Staffing and sampling
Jev changes the mix of work, not the existence of work. If auto-rate goes up, reviewer skill still matters — you are left with the ugly remainder. Sample autos every week; a silent criteria typo will look like “the model got worse.”
Do not use a Score to forecast handle time. Jaggedness: do not reconstruct magnitudes from Score levels. WFM tools own staffing math.
When 529s cluster, the runbook is “skip the smart gate,” not “page the whole company because AI is down” — unless the gate is the product.
Staff the remainder, not the vanished average
When auto-rate rises, reviewers inherit mixed language, missing fields, policy exceptions, and injection attempts. If you staff as if volume fell by the auto percentage, the queue will miss SLA on the tickets that actually need a human.
Pair weekly auto samples with reason codes from review templates and pages from the monitoring guide. Reviewer UI shows questions + probabilities, not a fake “Jev thought…” paragraph.
Product owns the action space; ops owns the default when the door is late. Write that default before go-live: usually review or page, never “call another model to guess the enum.”
529 and the skip-smart-gate runbook
When 529s cluster, the honest runbook is skip the smart gate: route to the pre-Jev queue, widen the HITL band, keep privileged actions fail-closed. Page the whole company only if the gate is the product (a Jev-shaped API you sell). Official SDKs retry 429/529 only; retries are not an incident plan.
Distinguish:
| Signal | Ops move |
|---|---|
| 401 | Key/door — not a model outage |
| 422 spike | Bad questions SHA — revert the pack |
| 429 sustained | Backoff + shed load; vendor limits can be dynamic |
| 529 / timeout | Skip-smart-gate; do not invent answers |
Review-rate jump + response.model change |
Alias move — pin jev-1.13.0 if floors exist |
Vendor “adding questions barely changes response time” is one-state fan-out, not a promise that a 10k backlog will drain faster.
When not to use this page’s advice
Do not publish AHT or CSAT lifts we did not measure. Do not let a Noul press an irreversible button. Do not treat SemIf/OpenJev latency as Jev capacity planning. Do not ask Jev to write the status page.
What this page does not claim
- No AHT claims.
FAQ
Can Jev set SLAs? No clocks. Score is not minutes.
Incident comms? Jev may classify; humans write status pages.
Should we page on every 529? Usually no — skip the smart gate. Page if the gate is customer-facing and you have no fallback.
Who owns weekly auto samples? Ops or QA with a named owner. Or the typo ships.
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.
Open-cluster pages are independent field-guide notes. Replicas and third-party interfaces mentioned anywhere on jev.pro are not Jev and not endorsed. Hub: Open. Siblings: monitoring guide, vs ticket queues, for product teams. Canonical: https://docs.typesafe.ai.