Community · E2 · artifact verified

Triage production alerts with four typed questions

Open-source alert triage where each production alert gets one Jev call with four typed questions - actionable, severity, team and paging - and auditable code turns the probabilities into routing; Jev never pages anyone, it only judges.

01 · Role in the system

What Jev does here

One call per alert asks four typed questions: a Noul for whether the alert is actionable, a Score for severity against published criteria, a Choice for the owning team, and the paging question. Plain code turns the answers into routes - and the demo's centerpiece is a guardrail: alert #5 shows P(page)=1.00 yet is still sent to human review, because code, not the model, decides who gets woken. The interactive demo replays a real jev-1.13.0 run over 8 staged alerts with latency and cost visible, and CI backs the repo. The design stance is explicit: probabilities in, auditable routing out, no model in the paging path.

02 · Control boundary

Where Jev sits

Jev as the judgment layer of on-call routing: one typed call per alert (actionable Noul, severity Score, team Choice), deterministic code converts probabilities into paging decisions, and the human-review guardrail lives outside the model.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

  • Staged demo of 8 alerts; no production traffic numbers yet.
  • Routing thresholds and team mapping are operator configuration.
  • Early project (2 stars at review time); docs are thorough but external validation is pending.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

  • mingleiw ↗Community · github · public · checked 2026-09-29