Community · E2 · artifact verified
Triage industrial faults from live telemetry
Fault triage for an industrial compressed-air unit with Jev: live telemetry in, classified alerts and tickets out, every judgment grounded in the unit's own service manual - read-only by design, with an eval harness that replays recorded decision cassettes.
01 · Role in the system
What Jev does here
Telemetry from the compressor streams into typed questions whose criteria come from the machine's manual, so a fault call is not vibes but a grounded reading: which fault class, how severe, what the ticket should say. The system is read-only by design - it classifies and drafts, it never actuates - and the eval side is built like a lab: decision cassettes with a JSON schema record judge answers so regressions replay deterministically, TypeScript tests cover the harness, and SPDX headers mark the whole thing CC-BY-4.0. Industrial OT meets a System One judge that only looks.
02 · Control boundary
Where Jev sits
Jev as the fault classifier for industrial telemetry: manual-grounded criteria judge live readings into alert classes and ticket drafts, the system stays read-only, and recorded cassettes make judge behavior replayable in CI.
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
- PoC scoped to one compressed-air unit and its manual; other equipment needs new criteria.
- 1 star; no field deployment or accuracy study against human technicians.
- Read-only by design - it informs tickets, it does not touch the machine.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- meddle-connect ↗Community · github · public · checked 2026-10-03