Know it was them. Judge it yourself.
Questions that stay in the test, a different paper for each candidate, and a record of every moment for you to review. Flags, never verdicts.
Free to start. No card.
- Questions
- One at a time
- Each candidate
- Their own set
- Records
- Camera, screen, mic
- Integrity score
- With its reasons
- Extra time
- Per candidate
- Rejects on its own
- Never
The problem
With Winkle
The problem
Questions get passed around
One photo of the screen and the test is in a group chat.
Questions arrive one at a time, lock if they leave, and go blank when another window is in front.
The problem
A chatbot can answer it
Paste the question in, paste the answer back.
Questions an AI can answer are flagged, with one click to replace them.
The problem
Verdicts nobody can check
A tool that marks someone down without saying why can't be trusted.
Flags with their moment, for a person to review. Nothing rejects on its own.
How it works in Winkle.
- 01
Choose your safeguards
Recording, leave-lock, the question pool or a length candidates choose, one device per link. Each is a switch, and most start on.
- 02
Candidates agree up front
They read rules written from your settings, and agree, before the first question.
- 03
Watch it live
See who is in the test, who left part-way, and when it submits itself.
- 04
You review
Every flag carries its moment. Jump to it, watch it, and decide what it means.
A question can’t leave the room.
Sent one at a time, locked if they leave, and gone the moment anything else comes in front.
One question at a time
The whole paper is never on the page. Each question is sent when they open it, so there is nothing to save in one go.
Leave, and it locks
Away longer than you allow, and every question they’d opened is submitted as it stands. Looking it up and coming back doesn’t work.
A clock on every question
Give each question its own time, or one per question type. The clock is kept on the server; when it runs out, what they saved counts and the question locks.
Blank when the window isn’t in front
Capture tools that take the screen get a blank, printing gives a blank page, and screenshot and print attempts are noted for you.
Full screen, when you want it
Switch it on and the test runs full screen. Stepping out covers the question until they return, and is noted.
Photograph a question for a chatbot, and the answer gives itself away.
Catch answers written by AI from a photo
Under every written and coding question sits a faint line meant for an AI, different for each candidate and each question. A person reading the screen passes over it. An AI handed a photo of it usually does what it says, and an answer that does is flagged for you to look at.
Questions an AI can’t answer from the wording alone
When Winkle drafts a test from its question bank, it reaches for questions an AI got wrong. Any it can answer are flagged on your Questions tab, with one click to swap them for harder ones.
Invented questions, an invented candidate and an invented hidden word. Illustrative — a rendering of the interface with invented data, not anyone’s record.
Every candidate sits their own test.
Different questions, in a different order, on a link that opens on one device.
Each candidate answers 15 of 40
Scored out of their own 15
Drawn per section, keeping the mix of question types and levels.
A different set for each candidate
Build a pool of 40 and ask each person 15. Each section draws its own share, keeping the mix of question types and levels, and everyone is scored out of the questions they were given.
Or let candidates choose how many
Set a range, say 10 to 50, and each candidate picks their length before they start. The time limit follows the number they pick, and results show the length next to the score.
Shuffled questions and options
Switch it on and each candidate gets their own order, so “the third one is B” helps nobody.
One device per link, on a computer if you say so
A test link opens in one browser. A second device is turned away and you’re told; if a laptop dies part-way, Reset device lets them carry on. Set it to Computer only and phones and tablets are stopped before the clock starts.
Three invented candidates and their draws. Illustrative — a rendering of the interface with invented data, not anyone’s record.
You see what actually happened.
Who’s in the test right now, who left part-way, and where two solutions are really one.
Live status, while they sit it
Active now, left mid-way or out of time, with how many they’ve answered, when they were last seen and when it submits itself. It keeps itself current; no reload.
Code similarity across candidates
Every coding answer is compared with other candidates’ answers to the same problem, across your company’s tests. Renamed variables, comments and spacing don’t hide a copy, and the starter code you gave doesn’t count. Matches from 85% open side by side.
Four invented candidates and two invented solutions. Illustrative — a rendering of the interface with invented data, not anyone’s record.
And every sitting, the way you’d review it.
The recording, the flags on its timeline, and the score with what each flag cost. Click a flag and you’re at that moment.
Integrity, on the record.
24 signals pinned to the second on the recording. You make the call.
Illustrative: the review screen with invented data, not anyone’s record.
Every signal, and what it means.
24 signals across tests and interviews. Each one is a moment to watch, not a finding.
Browser 10
- Left the tab
- Every time they leave the test, and for how long.
- Switched away from the window
- Another window came in front of the test, and for how long.
- Left full screen
- With full screen on, stepping out covers the question and is noted.
- Second screen detected
- A second display attached, in browsers that can tell (Chrome and Edge).
- Copied text
- Logged, and blocked on question text when copy-paste is off.
- Pasted text
- Logged with how much was pasted, and where.
- Text inserted by a script
- Words that appeared without anyone typing them.
- Automated browser
- A browser being driven by software, not a person.
- Tried to take a screenshot
- The screenshot key, caught where the computer lets the page see it.
- Tried to print
- Print or save as PDF, which only ever produces a blank page.
Session 3
- Tried to open the test on another device
- A second browser used the link and was turned away.
- Changed network
- They carried on from a different connection.
- Changed browser or device
- They switched machines part-way through.
Answers 6
- Answer followed an instruction hidden for AI
- The answer carries the word only an AI reading a photo would add.
- Answer matches another candidate
- Code that closely matches someone else’s solution to the same problem.
- Answer reads as AI-written
- A written answer that reads like a model wrote it.
- Tried to instruct the AI
- Text written to steer the grader rather than answer.
- Answered implausibly fast
- A long answer in less time than reading the question takes.
- Went past the time limit
- The clock is kept on the server, so it can’t be paused.
Camera and mic 5
- Face out of view
- Nobody in front of the camera.
- A second face in view
- Someone else in the frame.
- Looked away from the screen
- Sustained looking away, checked in the browser.
- Another voice heard
- A second person speaking during the sitting.
- A phone was held up to the camera
- Spotted in the candidate’s browser; no pictures are sent anywhere.
Never counted against anyone
Asking for help, being given extra time after a technical fault, a device reset by you, or a browser that can’t go full screen.
A score that shows its working.
A bare 72/100 asks you to trust a number. The same number with its reasons lets you check it.
- 1Every sitting starts at 100.
- 2Each flag takes points off, and names itself next to the points.
- 3Repeats count, so leaving the tab twice costs twice.
- 4Asking for help, or extra time after a technical fault, costs nothing.
- 5Nothing rejects anyone. The score is there to point you at the moment.
Integrity summary · one sitting
Example with invented data.
- 02:14Left the tab ×2-109s and 4s, both on question 2
- 07:40Pasted text ×1-10214 characters into the written answer
- 13:05A second face in view ×1-20Watch 13:05 to 13:20
Points differ between tests and interviews; the reasons are always shown.
Fair by design.
Hard to fake shouldn’t mean hard to sit. These are on the candidate’s side.
Blind grading
Answers arrive as Candidate 7. Names, emails and recordings stay hidden until a reviewer presses Reveal, which shows them to that reviewer only and is noted in your team’s activity log.
Extra time, for one candidate
For anyone who needs it. Every clock the test keeps stretches with it, section timers included.
Room for real life
Leave-lock waits out the grace you choose. Asking for help, or being given time back after a technical fault, never costs a point.
The rules, before the first question
Candidates read and agree to rules written from your settings, including how long they may step away.
Flags, never verdicts
Nothing rejects anyone on its own. A flag points you at a moment; a person decides.
An invented candidate and answer. Illustrative — a rendering of the interface with invented data, not anyone’s record.
Run it your way.
- What they see at the end
- A thank-you, pass or fail, or their score with feedback.
- Who hears when someone finishes
- You and the workspace owner, everyone who can view tests, or teammates you pick.
- How long invites stay open
- 3, 7, 14 or 30 days.
- Instructions in your words
- Shown before they start, and quoted in the invite when they’re short.
What’s included, and what isn’t.
Worth knowing before you choose it, rather than after. Proctoring comes with every test and interview.
Included
- Questions served one at a time, and locked when a candidate leaves
- Questions hidden when the window isn’t in front, and printing blocked
- A different set of questions for each candidate, or a length they choose
- One device per invite link, with a reset when you need it
- Live status while they sit it
- Code similarity across your candidates
- A faint line for AI on every written and coding question
- Questions checked against AI, with one-click replacement
- Time per question, full screen and computer only, when you want them
- Blind grading, and extra time per candidate
- Camera, screen and microphone recording, each a switch
- Face, gaze and phone checks, run in the candidate’s browser
- An integrity score that shows its reasons
- Recordings deleted 180 days after the decision
Not included
- Live human proctors watching each sitting
- ID document checks
- AI analysis of the recordings
- Reading emotions or personality
- Stopping a phone camera or an OS screenshot. No browser can
- Rejecting anyone automatically because of a flag
Does Winkle decide who broke the rules?
No. It notes what happened and when, as flags for you to review. Flags never reject anyone on their own.
Can a candidate screenshot the questions?
No browser can stop a screenshot, a screen recorder or a phone camera, so Winkle doesn't promise to. Questions go blank when another window comes in front, printing gives a blank page, and screenshot and print attempts are noted for you.
Can ChatGPT answer my test?
Winkle checks. Each question is put to an AI with the wording alone, and any it answers correctly is flagged on your Questions tab with a button to replace it. Written and coding questions also carry a faint line meant for an AI: an answer that follows it is flagged for you to review.
What about candidates who need more time?
Give one candidate extra time: +25%, +50%, +100% or a number of minutes. Every clock in the test stretches with it, and extra time never counts against them.
Are recordings analysed by AI?
No. Recordings are kept for people to watch. Face and gaze checks run in the candidate's browser and only raise flags; nothing reads emotions.
How is the integrity score worked out?
Every sitting starts at 100. Each flag takes points off and names itself, like "−20 · A second face in view", so you can check the number against the recording. Asking for help or being given extra time never costs anything.
Do you offer live human proctoring?
No. Winkle is built for hiring tests at volume, where a recording you can review beats a person watching live. There are no ID document checks either.
Start with this, free.
Questions that stay in the test, a different paper for each candidate, and a record of every moment for you to review. Flags, never verdicts.
Free to start. No card.