skip to content

For hiring teams

DevOps skills assessment on a real, broken cluster

You send a link. A live Kubernetes cluster with a genuine fault boots in under a minute, and the candidate gets VS Code and a terminal in a desktop browser and a time box of 25 to 45 min to bring it back. When they submit, grader scripts check the cluster. You get a report on how they got there and the full command log behind it.

Free plan, no card · Kubernetes only today · AI and search allowed

What it tests

Whether the cluster came back is one of 5 things it measures. The rest are about how. Every report scores the same criteria, 0 to 4 each, weighted into a score out of 100.

Fix correctness
Whether the fault is fixed, judged only by grader checks run outside the sandbox.
Diagnosis
Whether they followed evidence from the symptom to the cause before changing anything.
Blast radius
What their changes did beyond the fix: deleted state, removed guards, widened privileges.
Verification
Whether they checked their own fix before submitting.
Root-cause explanation
Whether they can explain the chain from cause to symptom, in their own notes.
Weights, and who scores what →

One lab, as the candidate sees it

A symptom, a short brief, and no hint at the cause. They get cluster-admin on their own cluster, so they can make it worse, and some of them do.

Getting the pod scheduled is half of this one. A fix that throws the data away to get there shows up under Blast radius, whatever the pod status says at the end.

The warehouse can’t schedule, and its data must survive

40 min
Advanced

The warehouse pod has been Pending since an infrastructure change, and the records on its volume must survive the fix.

What the hiring team gets

A report you can read without a terminal. Solved or Not solved, from the grader scripts and nothing else. A score out of 100 across the 5 criteria. One recommendation, from Advance down to Inconclusive.

Anything worth asking about in the next round is listed with the moment it happened, and the full command log sits underneath for whoever wants to check. Interview reports are kept 12 months.

It does not make the hire or no-hire call. You do.

Card payments time out, and ledger says it is healthy
Demo lab · not in either library
SolvedAdvance with reservations70/100· 22 of 45 min
Fix correctness3/4
Most of the weighted checks pass; at least one fails.
Diagnosis3/4
Followed evidence from symptom to cause. The wrong value appears in their own output before they changed anything.
Blast radius2/4
One major finding. Capped at 2 by the rules; the evaluator may only lower it.
Verification3/4
Checked the symptom, plus one check beyond it that the fix depends on or affects.
Root-cause explanation3/4
Explained the chain from cause to symptom.
Major · 14:02 · live edit on payments-api

They removed the liveness probe from payments-api directly, and the manifests still have it. Worth one targeted question.

Full command log and file diffs in the report

How it differs from a quiz

A quiz checks recall

It asks what a Pending pod means or which flag shows previous logs. Anyone with the docs open gets it right, and it says nothing about whether they can find out why this pod is Pending on this cluster.

This checks the cluster

There is nothing to recall. The fault is either fixed when they submit or it is not, and the command log shows the path they took. Search and AI are allowed, because the grader reads the cluster, not who typed the command.

A quiz cannot see the damage

Deleting the volume to get a pod scheduled is a wrong answer you can only give on a real system. Here it is scored, under Blast radius, even when the symptom clears.

Where each one fits

A quiz is cheaper per candidate and fine for sorting a few hundred applicants. This belongs at the stage where you are deciding whether to hire someone who will carry a pager.

Try it on yourself first

Free plan: 3 interviews and 3 trial runs on every interview lab, no card. After that, interview packs start at $350 for 10 interviews, not on sale yet.