Proving the Person: Hyrr's Full Interview Security & Verification Stack
An assessment is only as valuable as the certainty that the right person actually earned it. Here is every layer of identity verification, environment integrity, and recording review Hyrr runs before, during, and after an interview.
A skill assessment is only as valuable as the certainty that the person who earned it is the person who will show up on Monday.
That certainty has never been harder to establish. Remote interviews removed the physical room. Generative AI removed the difficulty of producing a confident-sounding answer to almost any technical question. A candidate today can be fed live answers by a second screen, coached by someone off-camera, or replaced entirely by a stand-in — and a conventional video call has no mechanism whatsoever to detect it.
Hyrr treats interview integrity as an architectural requirement, not a feature. Verification runs in three independent layers, each covering the gaps the others leave: proactive identity proof before the first question, environment lockdown in the browser during the conversation, and forensic reconciliation after it ends.
Here is the complete stack.
Layer 1: Proactive Identity Verification
On roles where identity verification is switched on, a candidate interviewing in the browser must prove they are who they claim to be before the first question. This happens once, and it is the anchor every later layer references.
Government-ID Verification and Selfie Matching
- Mobile handoff via QR code. The candidate scans a QR code from the interview lobby, which securely transfers the verification session to their phone — where camera quality and document capture are dramatically better than on a laptop webcam.
- Document capture through Stripe Identity. We capture high-resolution images of the front and back of a government-issued ID alongside a live selfie, using Stripe's document verification flow. Document authenticity checks — tampering, template mismatch, expiry — are handled at this stage.
- Selfie matching. Stripe Identity checks that the live selfie matches the photo on the ID, so the person who starts the interview is the person on the document.
Layer 2: Real-Time Environment Integrity
On roles where monitoring is switched on, the browser becomes a controlled environment once the interview begins. Three events count as strikes: leaving the interview tab, leaving full screen, and pressing the screenshot key. The first strike pauses the interview and puts an explicit warning in front of the candidate; the second ends the interview automatically.
The escalation is deliberate. A candidate who accidentally knocks their browser out of fullscreen deserves a warning, not a failed assessment. A candidate who does it repeatedly after being told is making a choice.
Everything else the browser watches — a lost window focus, a right-click, a blocked keyboard shortcut, a second screen — is noted on the report and never ends anything. Each of those also happens to honest candidates: a notification steals focus, a pop-up drops full screen for a moment, a shortcut is muscle memory. So they are recorded with a timestamp for the employer to see, and cost the candidate nothing on their own.
| Check | What it catches | What it costs |
|---|---|---|
| Leaving the interview tab | Alt-tabbing to a search engine, chat app, or AI assistant | Strike |
| Leaving full screen | Shrinking the window to make room for reference material | Strike |
| The screenshot key | Capturing a question to paste into another tool | Strike |
| Window focus | Another window taking the foreground | Noted |
| A second screen | A second display running a live transcription-and-answer tool | Noted; the candidate is asked to disconnect it before continuing |
| Right-click and blocked shortcuts | Context menus and system shortcuts used to copy questions out | Noted |
Multiple-monitor detection is worth calling out specifically. In browsers that support the Window Management API, such as Chrome and Edge, the platform detects secondary displays and asks the candidate to disconnect them before the interview continues. The event is noted on the report, not counted as a strike — a docked laptop is not evidence of anything.
Phone interviews are audio-only, so none of these checks applies to them; the report says so.
Why Looking Away Never Counts
Nothing watches the candidate's eyes live. The review of the recording afterwards may note that a candidate looked away from the screen, but that note is never scored and never ends an interview. Looking away, pop-ups and pauses never count as strikes. People look up to remember, a pop-up appears unasked, and a hard question deserves a moment's thought. Interrupting a candidate over an ambiguous signal would poison the assessment and punish nervous people for being nervous.
Layer 3: Post-Interview Forensic Assessment
After the session ends, the interview processor reconciles everything.
The reviewing model watches the recording alongside the full behavioral record and writes the report's Integrity headline: Clear, Needs a look, or Serious concern — with one sentence saying why, and its own account of what it saw.
Critically, the headline cannot contradict the record. Two strikes make it Serious concern whatever the model concluded. One strike, or a long silence the reviewer could not account for, makes it at least Needs a look — and so do the AI reviewer's own observations when it saw something worth a look. Every recorded event appears on the report with its timestamp and what it cost the candidate.
The record outranks the narrative because a browser event is a fact: if the candidate left the interview tab at 04:12, the answers reading well afterwards does not undo it. A long silence is held to a lower standard on purpose — it can have an innocent explanation, so it asks for a look rather than delivering a verdict.
Evidence, Not Verdicts
The employer's report lists what happened as plain rows — for example, Left the interview tab — Counted as a strike — Skip to 4:12 — with every row linked to the exact second of the interview recording.
This is the most important design decision in the entire stack: Hyrr does not make the hiring decision. Two strikes end an interview, but the platform surfaces the evidence and hands the decision to a human. A glance away might be a candidate checking handwritten notes they were told were permitted. A pause might be a candidate thinking. An automated system that failed candidates on signals like these would generate false accusations against real people, with real consequences for their careers.
So we show the employer what happened, timestamped and clickable, and let them watch the moment themselves.
Conclusion
No single check is unbeatable. Identity verification can be defeated by a sufficiently determined impersonator; browser monitoring cannot see a phone or a second computer; a review of the recording can only judge what the camera and microphone caught.
What is genuinely difficult is defeating all three layers simultaneously — passing the ID and selfie check, running with no second monitor, staying in the interview tab and in full screen, and holding up under a review of the recording, for the full duration of an interview.
That is the point. Layered verification does not make cheating impossible. It makes cheating harder than actually knowing the material.