Candidate A
- Reads the logs and documents before asking the AI
- Has the AI fix the cause, then runs the tests to prove it
- Rejects a change that breaks what already worked
- Explains in the Defense why this way
The hidden checks reward root-cause fixes
For organisations
Anyone can use AI. Not everyone can own the outcome.
See candidates' real work before the first interview
Have candidates take a mission from real work, with AI allowed. Hidden checks grade it, an AI reviewer asks why they fixed it that way, and the candidate shares the result with you: pass or not and a level for every piece, without an engineer reading code.
- No engineer has to grade it
- The same field for everyone
- People decide
Why now
The same submission does not mean the same person
AI makes output cheaper every day, so résumés, portfolios and take-homes all look about as good. What stays scarce is the person who knows what to ask the AI for, can check that what came back is right, and owns the outcome. That's what OperatorRank measures.
A made-up exampleThe same-looking submission
Candidate B
- Lets the AI patch each reported symptom
- The visible tests pass, so they submit
- Never notices the fix broke something that worked
- Can't explain it in the Defense
A fix that breaks what worked earns nothing
A résumé or the final code can't tell these two apart. The timeline, the hidden checks and the Defense can.
Hiring the old way
The three gates you screen with, AI now passes them all
Each gate used to tell you something about a candidate. Today AI helps everyone through them about equally well.
- Gate 1AI passes
A résumé shows they can write
AI makes every résumé look equally good. Teams filter by keyword and hope they didn't drop the right person; good engineers who write weak résumés fall out at the first gate.
- Gate 2AI passes
A take-home no longer shows who did it
AI finishes the whole assignment, submissions look alike, your team sees only the final result and spends engineer time reading code an AI wrote.
- Gate 3AI passes
Live interviews cost more and tell you less
Senior engineer time for every candidate, theory questions an AI answers better than anyone in the room, and in the end a gut feeling. One wrong hire is expensive.
Thousands of applicants
A thousand applicants. How much of your team's time?
Applicants arrive over your whole hiring window. On OperatorRank the hidden checks grade every one of them automatically, and your team opens only the reports worth reading. Move the slider and set the numbers to match your team.
- Read every résuméThe fastest old way, but it shows no real skill
33.3team hours
- Grade every take-homeAn engineer reads every one, and AI can do all of it
500team hours
- Live-code with everyoneSenior engineer time for everyone, a few candidates a day
1,000team hours
- A mission on OperatorRankHidden checks grade everyone; your team reads only the results worth reading
12.5team hours
40× less team time than grading every take-home, and 80× less than live-coding with everyone
Adjust your team's assumptions
The starting numbers are assumptions for comparison, not a promised result. Set them to match your team.
Fits the way you hire
Drop nothing. Put us into the flow you already have
Résumés, take-homes, interviews and live coding all stay. Put OperatorRank at any step: replace the steps AI can now do for candidates, or make it the last step. Try each one.
Before the résumé
Job post → OperatorRank → Résumé review → Interview / live coding → Decision
Ask everyone who applies to take a launch mission first and put the result link in the application, then open the résumés of people with strong evidence. Good engineers who write weak résumés stop falling out at the first gate.
Instead of the take-home
Job post → Résumé review → OperatorRank → Interview / live coding → Decision
Replace the assignment AI can do in full. The shared result shows, piece by piece, who really fixed the cause, without an engineer grading it.
Before the interview
Job post → Résumé review → Take-home → OperatorRank → Interview / live coding → Decision
Keep every step and add a mission before the interview round, then start your questions from the result the candidate shares.
Finish with us
Job post → OperatorRank → Decision
Make the mission the main gate. Your team sees everyone's real skill from the results they share. People on your team decide.
How it works
Three steps, from mission to decision
No tasks to write, no machines to set up for candidates, and nobody reading code.
Pick a mission and tell candidates
Choose one of the three launch missions and name it in your job post or a message to candidates; they take it on their own accounts.
They work with AI
60 to 75 minutes in a cloud workspace, with the same time and AI budget for everyone, and every action recorded on our server.
Read the result and decide
The candidate shares a link to the result: pass or not and a level for every piece. People on your team decide.
Candidates pay no money: practice uses the free credits on their own account.
A sample result page
Submit, and this is what you get
Between 531 and 816 hidden checks, depending on the mission, grade the work automatically, with a level from C1 to C5 for every piece of it. Nobody has to read the code.
After submitting, an AI reviewer asks why they did it that way; every launch mission requires it. Someone who let the AI do it blindly usually can't answer.
Your own people take the same launch missions on their own accounts and share the result with you: pass or not and the level for every piece, so you see where the team is strong. Companies see a candidate's result only when the candidate shares it.
Screens from OperatorRank · sample data
Between 531 and 816 hidden checks, depending on the mission, grade the work automatically, with a level from C1 to C5 for every piece of it. Nobody has to read the code.
Screens from OperatorRank · sample data
Why OperatorRank
What the usual ways of screening can't give you
Point by point, against the ways most teams screen engineers today.
| Keyword résumé screening | Online coding tests | Take-home graded by engineers |  OperatorRank | |
|---|---|---|---|---|
| What it measures | What the résumé says | Puzzles and algorithms | Depends on the task | A real system with a real problem, plus business documents |
| Can they use AI? | AI writes the résumé | Often banned or policed | Yes, but you can't see how | Fully, and we measure whether they use it well |
| Do you see how they work? | No | Often only the final answer | Only the final result | A recorded timeline the candidate can walk you through |
| Do they understand what they shipped? | Unknown | Unknown | Needs another call | The Defense asks everyone about their own work |
| How it's graded | Keywords that match the ad | Tests on standard puzzles | A reviewer's opinion, 30–60 min each | 531–816 hidden checks, the same set for everyone |
| Are candidates watched? | No | Often camera or screen recording | No | No camera, no screen recording |
| When do you get a result? | Instantly, with no skill shown | Instantly | Days | After the automatic check, with no reviewer to wait for |
| Who decides | A filter drops people by keyword | One overall score | A reviewer's opinion | People on your team, from evidence they can explain |
Compared with common ways of screening, not with any particular product.
Built for the AI era
Many tools still treat AI as cheating. We treat it as a work tool and measure who uses it well and owns the outcome.
See what the output can't show
Two people submit the same-looking work; the timeline and the Defense show who understood it and who accepted whatever the AI wrote.
Points for the root cause
Hidden checks reward root-cause fixes and give zero to a fix that breaks what already worked. Symptom patches score low.
No engineer time
Hidden checks grade the work automatically; your team reads only the results worth reading, even as thousands of applicants arrive.
Candidates want to take it
AI allowed like real work, no camera, nothing to install, nothing to pay. A good experience is your employer brand.
Made for Thai teams
The site, the results and the organisation app are in Thai and English, and candidates can hold the Defense in Thai. The workspace and the launch missions are in English.
In the organisation app
Everything your hiring team uses, in one place
Sample table of the level reached in each piece of the work Screen the first round without engineer time
Have candidates take a mission and share the result, let the system check it, and read each one: pass or not and the level for every piece. Nobody gets ranked for you.
Sample Defense conversation between the AI reviewer and the candidate Read how they think before the interview
The shared result shows where they're strong, piece by piece, and candidates can take you through their own Defense and timeline, so you walk into the interview with sharp questions instead of starting from zero.
Sample progress of a team assessment round Measure your own team, too
Your people take the launch missions on their own accounts and share the result with you, so the level for each piece shows where the team is strong and where it needs help.
Sample job board on the organisation side Post jobs in the same app
Open job posts and receive applications; a candidate can put a link to their own mission result in the application note, all in the same app where you read results.
Applications, interviews and offers
Move applicants through stages, message them with attachments, schedule interviews and send offers, all in the same app. No AI summary of who is better.
Connect your own AI through MCP
Create organisation tokens, choose what each one lets an AI reach, see its activity and revoke it at any time.
Missions on day one
Three missions from a real warehouse system
All three are Stockroom, a warehouse service that receives, allocates and picks stock. Candidates get the whole system, plus what the floor reported, the decision records and the runbooks. They work out what the real problems are, then get the AI to fix them right.
Stockroom Full
Deliver the signed receiving, stock-trace, picker and finance-close priorities in a warehouse service without breaking accepted records.
- 75 min
- 816 hidden checks
- Defense required
Stockroom Focused
Repair allocation-ledger, reservation and ASN import behavior in the warehouse service.
- 60 min
- 655 hidden checks
- Defense required
Stockroom Receiving
Repair allocation-ledger, ASN import and expiry/FEFO behavior in the warehouse service.
- 60 min
- 531 hidden checks
- Defense required
Mission content is in English · Hidden checks counted on 19 Oct 2026
What the hidden checks reward
Patching symptoms
- Quick one-line patches for each report
- The visible tests may all pass
- Break something that worked and that fix earns nothing
Scores low
Fixing the cause
- Reads the documents and works out which rule is right
- Has the AI fix the cause, then checks it really did
- Doesn't break what already worked
Points sit on root-cause fixes
Why you can trust the result
A test with a clear spec, AI can pass for you.Our missions need a human head.
Real work never comes with a spec. It comes as a symptom, scattered evidence and business rules that sometimes disagree. The people worth hiring find what's right, then get the AI to do it right. That's what we measure.
- 01
Taken from real work, with no spec to copy
Candidates get a real system with a real problem, plus the business documents and what customers reported. The code shows what is wrong; the right answer is in the business evidence (docs, logs and tickets), not in a hint.
What candidates get at the start brief.mdWhat happenedops/complaints.mdWhat the floor reporteddocs/adr/Decision recordsdocs/runbooks/Runbookslogs/System logsstockroom/The system's codetests/The visible tests
- 02
Use AI freely, but everyone gets the same
A cloud workspace with AI built in. Everyone gets the same time and the same AI budget and chooses from the same set of models, and every action is recorded on our server, not sent in by the candidate.
The same for everyone Time60–75 minutes by mission, kept by the serverAI budgetThe same for everyone; when it runs out, the AI stopsModelsThe same set, chosen by the missionHidden checksThe same set for everyoneRecordedEvery command, question to the AI and code change, on our server
- 03
Graded by hidden checks the candidate never sees
Between 531 and 816 checks, depending on the mission, the same set for everyone. Points sit on the real root-cause fixes, not on quick one-line patches; break something that already worked and that fix earns nothing. Per-check results are kept for review or appeal.
Sample table of the level reached in each piece of the work Sample result page: passed, every piece reached its required level
The candidate experience
Good candidates want to take this
The assessment is a candidate's first real contact with your company. Strong people don't want puzzles or to be watched; they want real work, real tools and to see their own result.
- AI allowed, like at workNo sneaking, no guessing what's forbidden
- No camera, no screen recordingOnly the work in the workspace is recorded
- Nothing to install or payIt runs in the browser, on the free credits in their own account
- They see their own resultWhat passed, what didn't, and a Defense in their own language
Fairness and privacy
The same field for everyone
What's recorded
- Commands, files, diffs and tests run in the workspace
- The conversation with the AI and the AI budget used
- The whole Defense conversation
What's never collected
- Camera or microphone
- Screen recording
- Anything typed or opened outside the workspace
Same time
The clock runs on the server. Closing the browser doesn't stop it, and nobody can add time.
Same AI choices, same budget
The mission sets which models can be used and everyone chooses from that set. The AI key never sits in the workspace.
A workspace of your own
No outbound internet, and nothing you do there touches anyone else.
No surveillance
No camera, no screen recording, nothing outside the workspace.
If we break, you can appeal
Outages are on the record. You get credit back, a retry or a re-check, as the policy says.
People decide
No AI hires or rejects anyone. The system gives evidence; a person reads it and decides.
Schools and courses
Teach with it, too
Schools, universities and training courses open classes, keep student lists and post announcements in one place; students practise on the launch missions with their own accounts.
- Classes, terms and student lists
- Announcements to the class
- Students practise on their own accounts
- Results are learning progress, not hiring evidence
If candidates can use AI, what are you measuring?
Whether they can find what's right. Missions come with no spec to follow; the right answer is in the business evidence. Points sit on the real root-cause fixes, and a fix that breaks something that already worked earns nothing.
Do we have to drop résumés or our interviews?
No. Put OperatorRank at any step: before the résumé, instead of the take-home, before the interview, or as the last step. CVs stay open as usual; everyone just gets to show their skill first.
A thousand applicants: what extra work is that for our team?
No grading. Hidden checks grade everyone automatically, and your team opens only the results worth reading, so its time goes to the people worth talking to, not to everyone who applied.
How is this different from online coding tests?
Most tests use puzzles and often ban AI or watch candidates on camera. Here candidates fix a real system with AI fully allowed, hidden checks reward fixing the root cause, the timeline records their work, and they explain it in the Defense, with no camera and no screen recording.
How fast are results?
There's no reviewer to wait for: hidden checks grade the work automatically after a candidate submits. Where a mission requires the Defense, the result shows once the candidate has finished it.
What does an organisation see of a candidate?
Only what the candidate shares: a result's share link shows pass or not and the level for each piece of the work.
Do candidates pay?
No money. Candidates take the launch missions with the free credits on their own account, and there's nothing to buy.
Does the system decide who gets hired?
No. The system gives evidence; people in the organisation read it and decide. There is no overall score that ranks people, and no automatic hire or reject button.
Can we measure our own team?
Yes, with the launch missions: each person takes one on their own account and shares the result link with you, which shows pass or not and the level for each piece.
Can we post jobs here?
Yes. Open job posts and receive applications in the same app where you read results; a candidate can put a link to their own mission result in the application note.
Can it connect to our own AI?
Yes, through MCP with organisation tokens. When you create a token you choose what an AI can reach, and you can revoke it at any time.
Use AI. Own the outcome.
Anyone can use AI. Not everyone can own the outcome.
Start hiring on real work. The first missions open on 30 October.