Hire Ops engineersunder real failure
Real infrastructure, real load, hidden faults. Add a candidate's email and get the exam, every step on record, and a report with interview questions.
Candidate view. The candidate gets an exam invitation by email, opens the link, reads the brief and starts the exam. While ticket traffic surges on Unicorn Tickets, a concert booking service, the exam AI suggests scaling out; the candidate holds it because CPU has headroom and finds connection pool exhaustion in the logs.
Still measuring operations with coding tests?
The ability to hold up under a real incident isn't measured by a coding problem. Doing it properly takes a senior's time to build the environment and grade it — and that time is always short. Worse, today's take-home answers can just be written by AI.
- What it measures
- Typical take-home: Design Q&A, algorithm problems
- Paged examResponse under live incident load
- Who prepares it
- Typical take-home: A senior writes it, builds the environment, grades it
- Paged examAdd an email — environment and grading are built in
- AI use
- Typical take-home: Banned, or effectively unenforceable
- Paged examAllowed, and how it's used is recorded and scored
- Copying
- Typical take-home: An answer can be copied verbatim
- Paged examA system that holds under load can't be copied
Add an email. Get a report and interview questions.
Add a candidate's email and an invitation goes out with the exam brief and a start link.
Real infrastructure takes real load and a hidden fault. The candidate works it for 30 minutes with a terminal and an AI assistant.
Every command, every AI exchange, every live metric lands in an events table you can replay later.
An evidence-backed report and interview questions tailored to that candidate are generated automatically.
We don't ban AI. We watch how it's used.
Candidates get an AI assistant with a test key and a budget. Every prompt is recorded, and scoring checks whether the candidate verified the AI's reasoning, blocked a wrong action, or approved the right one. Some scenarios set an AI trap where the obvious fix actually makes things worse.
Hidden fault injection
The root cause sits inside the infrastructure for the candidate to find.
A cause and numbers that change every time
Each candidate gets different numbers and details, so answers can't be copied.
Scored on AI collaboration
Scoring checks whether the candidate verified the AI's judgment, not whether they used AI.
AI trap scenarios
Scenarios where the plausible AI fix makes things worse — and whether the candidate catches it.
The four questions we hear first
- Won't this take even more of our senior engineers' time?
- Paged drafts the exam, the environment, the grading, and the interview questions. Your senior just reads the report and runs the interview.
- How do you stop cheating?
- Each candidate gets a different cause and different numbers, and every command and AI exchange is logged. Copying an answer won't revive a system that's failing right now.
- Is our cloud account at risk?
- Exams run on a dedicated test account or in the Paged sandbox, never on production — with a budget cap and automatic teardown when the exam ends.
- Can we run this at campus-hiring scale?
- Each candidate gets their own isolated environment. We'll size it with you on a demo call.
After the hire, train on the same incidents
New hires are ready for on-call sooner, and your team's own postmortems become the next drill.
Postmortem
The team records a real incident and marks the symptom, the trap, and the cause.
On-call drill
That postmortem becomes a drill the team works through together.
Graduation exercise
Confirms a team member is ready to go on-call for real.
Team readiness
Completion builds into a picture of the whole team's readiness.
Team readiness feeds the next job description.
No comparing people with each other. Never used for performance reviews.
Coming soon · Verify — put AI SRE agents through the same incident and compare them before you adopt one. Explore Verify →
Exams run outside production. People make the call.
- The final decision is always made by a person.
- Every score comes with its evidence.
- Data is retained and deleted according to regional rules.
- Separate test account
- BYOC is recommended; nothing is installed on your production account.
- Paged sandbox
- Start right away with nothing to install.
- Budget and time caps
- A budget cap, a time limit, and automatic teardown when the exam ends.
- Fault layer hidden from candidates
- The layer that creates the fault is never visible to candidates.
- Seoul region storage
- Data for Korean customers is stored in the Seoul region.
- Retention you control
- You set the retention period; data is purged automatically after it.
FAQ
Can candidates use AI?
Yes. AI use isn't banned, it's part of what's scored — what they asked for, and how they verified the suggestion.
What gets recorded?
Commands run, AI prompts and replies, and live metrics. No screen recording or face matching.
How long is data kept?
You set the retention period. Data is purged automatically once it passes.
What do you install in our account?
Only an outbound-only agent, in a dedicated test account — never in production. To start with nothing installed, use the Paged sandbox.
Which roles can this evaluate?
Operations-heavy roles like DevOps, SRE, DBA, and SA.
Can we try it for free?
Yes. A free trial lets you build an exam in the sandbox with nothing to install.
Make Ops hiring real
Answers can be copied. A system that holds under load can't.