We don't list certifications we haven't earned. Instead, this page explains — plainly — how Skillytal's architecture actually handles proctoring signals, assessment data, and AI-assisted review.
Every automated signal Skillytal produces exists for one purpose: to help a human reviewer make a better decision. It is never the decision itself.
Proctoring and integrity outputs are surfaced as signals for a person to review — never as an automated pass/fail or accusation.
Camera, audio, and timing observations are combined and weighed together. No one input — on its own — determines an outcome.
The system generates signals. Hiring teams and reviewers make decisions — with full context Skillytal can't see or judge.
What this means in practice: Skillytal does not claim 100% cheat detection. It does not perform lie detection or claim to read emotion or deception from a candidate's voice. It does not guarantee detection of every AI overlay, screen-share tool, or integrity risk — sophisticated attempts to circumvent any proctoring system can and do exist. And it never issues a fully-automated rejection based on a proctoring signal alone.
If that sounds like a smaller promise than competitors make, that's deliberate — we'd rather be accurate than impressive.
No certification badges — just how data actually moves through the product.
Where the architecture supports it, proctoring signals are computed on the candidate's own device, keeping raw video processing close to the source rather than streamed by default.
The product is built around minimizing what needs to leave the candidate's device or organization in the first place, not maximizing what gets collected.
Assessment links are scoped to a single candidate and time-bound to the session, reducing the window in which a link could be misused or shared.
Every organization enables only the question formats and proctoring checks a given role actually needs — nothing runs by default that wasn't turned on.
Assessment records and reports are scoped to the organizations and roles that need them — a candidate's data doesn't surface outside that context.
AI-generated signals are structured for human review, complete with context — not surfaced as an unexplained score with no way to inspect it.
Built to scale across teams, roles, and hiring volumes without requiring a re-architecture as an organization grows.
Where the product architecture supports it, the first pass of proctoring analysis happens directly on the candidate's device — before anything is sent anywhere.
Frames are analyzed locally rather than streamed continuously for processing elsewhere
What leaves the device is a structured signal, not an unbounded video stream
Flagged moments are packaged for a human reviewer to look at in context
This describes the general architecture pattern Skillytal is built around, not a guarantee for every deployment or device configuration.
Being clear about the limits of AI-assisted proctoring is part of using it responsibly.
No proctoring system can catch every possible way to circumvent an assessment, and Skillytal doesn't claim otherwise.
Audio analysis looks at communication and speech signals — it never claims to detect deception or emotional state.
Signals are framed as things to review, not statements of fact about what a candidate did.
Skillytal doesn't reject or disqualify a candidate on its own — that decision stays with the hiring team.
Where the architecture computes proctoring signals on-device, raw video isn't the primary artifact that leaves the candidate's session by default. Organizations can configure their assessment policies to match what a given role requires.
Proctoring signals and flags are scoped to the hiring team and organization running that assessment — they aren't shared across unrelated organizations or roles.
Yes. Assessment policies are configurable per role, so a team can enable only the question formats and integrity checks that make sense for a given position.
No inflated claims, no unverified badges — just an honest account of how Skillytal handles proctoring, data, and AI-assisted review.