Strictness is not one dial. A fully proctored CodeSignal session can run six independent monitoring channels, each switched on separately by whoever built the test, and each recording something narrower than candidates assume. That leads straight to the thing most people get wrong: a setting that hides a window from screen capture is relevant to exactly one of the six. It has no bearing on the camera, none on the microphone, and none on the identity check.
Variance matters as much as strictness. A Certified Assessment, whose score you forward to several employers, is proctored heavily because it must be trustworthy to a company that never watched you take it. One employer's own test is configured from a settings panel, and plenty go out with almost nothing on. The honest answer to "how strict" is "count the channels enabled on yours".
The six channels, and what each one really watches
Read the last column first. It is the question most people arrived with.
| Channel | What it literally captures | How strict in practice | Does hiding a window from screen capture matter here? |
|---|---|---|---|
| Full desktop recording | Video or periodic frames of the display you granted. A strict configuration asks for the whole screen rather than one tab, so everything on it is in frame. | Strict where present, but never silent. The browser shows an explicit share prompt naming what will be shared, and keeps a sharing indicator up all session. | Yes. This is the one channel of the six that a capture-exclusion setting has any bearing on at all. |
| Webcam | Stills at intervals, or continuous footage in some configurations, of whoever sits in the camera's cone. | Strict on Certified Assessments, absent on many company-built tests. A reviewer notices a second person, an empty chair, or sustained glances off to one side. | No. The camera points at your face, not at your monitor. Nothing you change about what your display shows can appear in that footage. |
| Microphone | Room audio for the session, where audio capture is enabled. The browser asks for microphone permission separately from camera permission. | The least commonly enabled of the six, and the hardest to explain away when it does catch something. The signal is a second voice or one half of a conversation, not the code. | No. Audio is captured from the air in your room. A screen setting is not in that path in any sense. |
| Government-ID photo match | A webcam photo of your face and, in flows that require it, a photo of a government ID held up to the camera. Both are stored against the session. | Blocking. It runs before the timer and you cannot start without completing it. Its job is attribution, so a portable score belongs to a named person. | No. It happens before the assessment opens, through the camera, and the record it creates has nothing to do with your screen. |
| Paste events | That content arrived in the editor as one event rather than as typing, with a timestamp, and in many configurations the pasted text itself. | Logged reliably, interpreted loosely. Pasting a helper you wrote yourself in task one looks identical to pasting from anywhere else until a human reads both. | No. The event is recorded by the assessment page inside your browser at the moment of the paste, independently of what any recording shows. |
| Typing rhythm and code replay | The build-up of your solution as it was written, so a reviewer can play it back rather than only read the final state. | The most underestimated of the six. A solution that arrives whole, in order, with no failed run and no exploratory print reads differently from one that was fought for. | No. Replay is reconstructed from what reached the editor. The screen it was typed on is not part of the record. |
Alongside these sits full-screen enforcement, a browser mechanism rather than a monitoring channel. A proctored assessment pushes the tab into full screen and logs each exit with a timestamp and a duration. It records that focus left, not what you switched to. It cannot name the application you opened, read another tab, or see a second monitor.
Five of the six do not involve your screen at all
The assumption underneath this query is usually that one clever display setting neutralises proctoring. It does not, and the reason is physical rather than technical. Capture exclusion changes what a screen-capture stream contains. Four of the other five never touch such a stream: the camera samples light off your face, the microphone samples air, the ID photo was filed before the clock started, and a paste event is a note the editor made about its own contents. The fifth, code replay, is assembled from keystrokes, so it reads identically whether your monitor was recorded, mirrored, or switched off.
What escalation actually looks like on one timeline
Single flags are common and mostly noise. Reviewers escalate on correlation, because correlation is the only thing that survives an innocent explanation:
Twenty-two minutes on task two with three failed runs. Task three then closed in four. At minute fifty-one, one paste of roughly forty lines. The replay for task three shows the solution arriving top to bottom in near-final form: no failed run, no print statement, no variable renamed. Across the same four minutes the webcam stills show the candidate looking down and to the left rather than at the screen.
Each is defensible alone. The paste was your own utility. Fast finishes happen on a problem you have seen. People look away when they think. Four channels agreeing on one four-minute window is what gets a human asked to look, and even then the usual outcome is a request to walk through the code live rather than a silent rejection. Which is the practical argument for understanding a solution rather than possessing one.
Where OfferPilot AI is the wrong tool
Being direct, since a vendor publishes this page. Code Copilot is a Windows 10 and 11 desktop overlay that reads a problem off your screen and returns a solution with the intuition, a dry run and the Big-O analysis. Its window is excluded from screen capture at the operating-system level, so it is absent from Zoom, Teams, Meet, OBS and screenshots while staying visible on your own monitor. Set that against the table above and the limit is obvious: it addresses one channel out of six.
- It does nothing about a webcam, a microphone, or an ID check. Nor about a lockdown environment, an invigilator in the room, eye tracking, or a phone propped up as a second camera. None of those are screen capture.
- It does nothing about the logs inside the assessment page. Paste events, focus events, timings and code replay are recorded by the test itself, and no display setting reaches them.
- A proctored Certified Assessment is the clearest case to leave it out of entirely. The value of that score is that a company which never watched you earn it can rely on it.
- It is Windows only. No macOS, no Linux, no mobile, no browser extension.
The use that holds up is before the assessment window opens. Working timed problems against something that explains why an approach works is what makes a replay read like fluency instead of recall. For the interview that follows a passing score, the company Q&A generator builds a question bank for that firm and role, and the question library covers the rounds after the screen.
Counting your own channels in sixty seconds
- Does the invite say "Certified Assessment"? If so, assume the full set is on.
- Did setup ask for a photo of your face or an ID? That is channel four, already complete before you code.
- Did the browser ask for camera permission, and separately for the microphone? No prompt, no capture on that channel.
- Did a share prompt appear naming a screen or a tab? That, and only that, is desktop recording.
- Assume paste logging and code replay are running regardless. They need no permission and produce no prompt.
Keep that last line. The two channels that never announce themselves are the two candidates forget, and they carry the most weight when a reviewer opens the report.
Related guides
- Does HackerEarth Use Your Camera? Every Proctoring Toggle
- Does HackerRank Record Your Screen? Proctoring Tiers Explained
- Can HackerRank Detect Copy and Paste? What the Report Shows
- Codility Similarity Check Flagged: What Triggers It and Why
- What Recruiters See in Your HackerRank Report: Field by Field