Practice real infrastructure, not theory.
Real outages. Real terminal. Graded on your approach.
Every request is failing. Restarting the app won't fix it — can you see why?
this week's incident · #24 · new every Monday
Every data request returns 500 — but the app is healthy.
Two ways to be graded
Every problem runs in one of two modes.
Mode 01 · Execute
Fix it in the terminal
A live lab terminal plus a task checklist. Run real commands against the broken system — your transcript and the final state go to the judge.
Mode 02 · Diagnose
Name the root cause
Explore the broken environment, then write up the root cause and the fix. Judged on your investigation and your diagnosis — the reflex you need at 3 AM.
What happens when you hit enter
No cloud sandbox to boot. No fabricated output.
Your command
Real commands in a browser terminal — docker, kubectl, systemctl. Nothing to install or spin up. Stuck? Type hint.
The Karma engine replays the incident
Every incident is captured from real systems and replayed deterministically. Same command, same result, every time — no environment that drifts from reality.
The judge reads your transcript
An AI judge scores your whole approach — the commands you ran, the wrong turns you avoided — not just the final state.
✓ root cause named · ✓ fix verified · approach scored
We built this because nothing else worked.
Cloud sandboxes cost money, take forever to spin up, and drift from reality. OpsKarma is free while in beta — no card, no plan to choose. We'll be honest about it if that ever changes.