What a flag means — and what it does not

Proctoring products have a habit of turning observations into verdicts. A candidate looks away; the software says “suspicious”; a recruiter under time pressure reads that as “cheated”. We built Clevid so that chain cannot form.

What the client observes

During a session the client records three kinds of things, all timestamped against the server clock:

• Environment findings from the preflight check: virtual machine, remote session, virtual camera or audio device, multiple displays.
• Events during the test: the test window losing focus, a new application opening, the recording pausing.
• Continuous webcam video and microphone audio, uploaded in short segments.

What a flag is

A flag is an event, or a combination of events, that a reviewer should look at. Focus leaving the window for three seconds is a flag. An unauthorized app opening during the test is a flag. A gap in the recording is a flag.

A flag is not a score change, a verdict, or a probability of cheating. We do not compute a “cheating score” because we do not believe one can be computed honestly from software signals alone.

What the reviewer sees

Each flag sits on a timeline. Clicking it opens the recording at that exact second. The reviewer can see whether “focus left the window” was a low-battery popup or a second application, because the video is right there.

The reviewer then does one of three things: dismisses the flag with a note, requests a retake, or records a decision. All three are audited with who and when. A second reviewer cannot silently overwrite the first.

Missing evidence is stated, not scored

If the webcam recording does not cover the whole attempt — the machine slept, the upload failed — the report says so: coverage incomplete, 38 of 45 minutes. It does not lower the score, and it does not mark the candidate as suspicious. A result is marked Verified only when coverage is complete; otherwise it is marked exactly what it is.

Why this matters to companies, not just candidates

A hiring decision you may one day have to defend — to a candidate, to a manager, to a regulator — needs to be explainable. “The software gave them a 0.7” is not an explanation. “A reviewer watched the footage at 12:04, saw a battery popup, and dismissed the flag” is.

Read more on the How it works page, including the list of things no software can catch.