How It Works
Three layers. Each honest about what it covers.
Software locks the machine. The camera watches the room. A person decides what a flag means. Clevid is built so that each layer covers the gaps of the one before it — and so that we never have to claim more than we can prove.
- Start a hiring roundStart free (10 attempts / mo) →
- Transparency & boundariesRead our honest limits ↓
- Candidate experienceWhat candidates see →
01 · Defense in Depth
Machine lockdown, room evidence, and human judgment.
No single control is sufficient on its own. Combining a protected desktop window, continuous recording, and an audited human review queue ensures every score is defensible.
Dedicated desktop lockdown
The client keeps the test window in front and blocks other applications, floating overlays, screen sharing, and the clipboard for the duration. It runs as a standard desktop application — no permanent system drivers, nothing that stays running after the exam.
- Preflight refuses virtual machines, remote desktop sessions, virtual cameras and audio devices, and multiple displays
- Focus changes and new applications are logged with timestamps
- Server-side deadlines and answer saving; the candidate’s clock is never trusted
Continuous recording & timeline
From Start to submit, the webcam and microphone record the whole attempt and upload in verified segments. The candidate cannot pause or stop it. A result is marked Verified only when coverage is complete; any gap is stated on the report rather than turned into a penalty.
- Every event on the timeline maps to the exact second of the recording
- Segments are integrity-verified before they count as evidence
- Every view of a recording is logged, because it is biometric data
Human review and appeals
Nothing is decided by software. Flags wait in a review queue with the footage attached. Your reviewer dismisses the flag, requests a retake, or records a decision; the candidate can appeal through your team.
- Reviewer assignment and stale-review protection so two people never overwrite each other
- Retakes issue a fresh invite and prefer questions the candidate has not seen
- Every decision is audited with who, when, and why
02 · Assessment Lifecycle
From publish to decision in five steps
Build
Pick a subject and level mix from the bank or add your own questions. Publish freezes the paper.
Invite
Paste emails, upload a CSV, or sync Workday. Each candidate gets a one-time code; delivery status is visible.
Preflight & notice
The client checks the machine and states what will be recorded. A blocking issue never spends the invite.
Locked session
Lockdown engages, the webcam and microphone record, answers save as they go. Submit releases everything and the client stops.
Review
Scores are graded server-side and ranked. Flags wait for a person. Export when you are done.
03 · Candidate Report
What is in the integrity report.
One page per candidate. Every line links directly to the underlying timestamped evidence.
- 01Score and cohort rankPoints per section, competition rank across every invited candidate.
- 02Recording coverageComplete, incomplete, or failed — and the exact minutes covered.
- 03Environment verificationVM, remote-session, virtual-device and display checks with their results.
- 04Event timelineEvery focus change and process start, mapped to the exact second of recording.
- 05Reviewer decisionsWho looked, what they decided, and the audit note they left.
- 06Notice versionWhich recording notice the candidate saw, and when they started.
04 · Honest Limits
What software cannot catch — and what we do about it.
We say “dramatically reduces and detects”. We never say “impossible”. Full lockdown ships on Windows 10 and 11 today; macOS is in progress and will ship with a clearly labelled assurance tier.
| Scenario | Beyond Any Software Lockdown | How Clevid Handles It |
|---|---|---|
| Off-camera secondary device | A phone or tablet propped up outside the camera frame cannot be detected by local process checks. | Covered by continuous webcam and microphone recording plus human review — never an automated AI accusation. |
| Hardware audio & room assistance | A Bluetooth earbud, hardware capture splitter, or pre-briefed accomplice in the room bypasses OS window checks. | Continuous room audio and video recording gives human reviewers full environmental context. |
| Question set memorization | A determined candidate sharing static questions with future test-takers. | Per-candidate papers are drawn dynamically from subject banks by difficulty tier so leaked questions have minimal value. |
| False-positive hardware events | An OS battery notification or background update briefly stealing window focus. | Zero automatic fails by design. Reviewers jump to the exact video second, see what happened, and can issue a retake. |