Integrity & Proctoring
Integrity in layers. Judgment by people.
Choose the safeguards and capture modes that fit each assessment, connect findings to their source moments, and give reviewers the context they need to confirm, dismiss, or explain what happened.
Signals inform review. A named person owns the resolution and outcome.
Five layers, each sized to the stakes
Browser-native safeguards — nothing to install.
Configure supported browser controls such as fullscreen policy, clipboard restrictions, print-attempt guards, screen-share continuity, and event logging without requiring a separate candidate application.
- Fullscreen enforcement with per-assessment policy
- Keyboard, clipboard and print-attempt guards within supported browsers
- Second-display block at entry and mid-exam detection
- Stopping screen share locks the exam until sharing resumes
- Virtual-machine signals, DevTools attempts and copy/paste events
- A session watermark (candidate + timestamp) deters leaked screenshots
Consented capture, watched live if you choose.
Webcam, screen and microphone capture — each an explicit per-assessment choice with candidate consent named up front. Authors pick photo or video modes and the snapshot cadence; a live monitor wall shows active sessions as they run.
- Webcam: periodic photos or continuous video — your choice
- Screen: screenshots or full recording, with author-set intervals
- Microphone capture (off by default, consent named on the candidate screen)
- Live monitor wall across active attempts
- Per-workspace retention windows and per-attempt erase
One-to-one signal. Named human decision.
Show the session selfie, photo ID and optional reference photo together. Where enabled, a one-to-one similarity result can help a reviewer compare the submitted face images without searching a gallery or other candidates.
- Session selfie and photo ID capture
- Optional submitted reference photo
- One-to-one similarity result when enabled
- No gallery or cross-candidate search
- Named reviewer records the identity decision
Connect integrity events to their source.
Match-based signals, correlated events, cohort patterns, and a hash-chained timeline help reviewers inspect what happened without treating one event as proof of misconduct.
- Honeytoken matches when protected question text returns in an answer
- Hash-chained event log with chain status available for audit
- Large-paste detection in written answers; synthetic typing-speed detection
- Correlated findings: focus-loss followed by a paste, stitched into one story
- Collusion analysis across candidates: identical wrong answers, near-duplicate essays
- Tab switches, fullscreen exits and connection drops logged as context
AI points to moments. A reviewer resolves the finding.
When requested in the review workflow, AI can inspect captured media and surface moments worth opening. Reviewers see the source and available scan coverage before recording what the finding means.
- Webcam frames: absence, extra people, phones, headsets, notes, gaze-off, covered lens and more
- Screen captures: AI-assistant windows, search engines, messaging apps, remote-control tools
- Audio transcription with flagged speech segments — without claiming speaker identity
- Available scan coverage disclosed to the reviewer
- A named reviewer records finding resolutions
- Run metadata and model details retained for provenance
Proportional security, sized to the stakes.
Not every assessment needs every layer. A quick screening might log basic signals; a certification final can run the full stack — secure mode, recording, forensics and AI review. You choose per assessment; candidates see exactly what applies before they consent.
Whatever the level, signals and AI remain review context. They do not independently set a score, disqualify a candidate, or record the final decision.
Security & complianceSet the level
Pick the layers this assessment's stakes deserve.
Consent first
Candidates see every active control before starting.
Capture & chain
Evidence is recorded and hash-chained as it happens.
AI triage
On review, AI flags the moments worth opening.
Reviewer resolution
A named person confirms, dismisses, or adds context.
The stack at a glance
Browser Safeguards
Configure fullscreen, clipboard, print-attempt, screen-share, and activity controls.
Live Monitor Wall
Watch active sessions live — webcam and screen frames as they arrive.
Honeytokens
Surface a match when protected question text returns in a submitted answer.
Hash-Chained Event Log
Chain session events so edits, deletions, or reordering can be detected.
Collusion Analysis
Identical wrong answers and near-duplicate essays surface across the cohort.
Human-Reviewed Findings
AI can surface webcam, screen and audio moments with their source context.
Identity Review
Use submitted identity materials and an optional one-to-one signal; a named person decides.
Audit Pack
Integrity grade, chain verification and the full timeline, exportable per candidate.
Reports & exportsLockdown alone is not integrity.
Lockdown-only tools
- A single control profile for assessments with different stakes
- Signals can be treated as outcomes without enough context
- No evidence trail a candidate could contest
- AI verdicts with no human in the loop
SkillCort Integrity & Proctoring
- Configurable browser safeguards and capture modes per assessment
- Evidence-linked signals: honeytokens, hash-chained logs, and cohort patterns
- AI findings retain source context for named human resolution
- Consent-first capture with disclosed controls and retention windows
- An audit pack that makes every conclusion checkable
The strongest lockdown still can't replace evidence and judgment. SkillCort gives you all three, in proportion.
What we will not build.
Integrity exists to make evidence defensible — not to surveil people. Some capabilities are excluded on principle, and the exclusions are published, not just promised.
- No emotion or personality inference from video or audio
- No covert behavioral profiling
- No automated face matching, gallery search or stored face template — a session selfie and photo ID are compared side by side by a named human reviewer
- No cross-client candidate memory or hidden reputation scores
- Integrity signals do not independently disqualify candidates
- Reviewer resolutions and decision ownership retained in the audit trail
Frequently asked questions
No. Signals — from a tab switch to an AI flag — are context for a human reviewer. Nothing fails automatically; the one configurable exception is the secure-mode violation limit, which an organization can explicitly set to auto-submit an attempt, and that too is a submission for review, not a verdict.
No. AI review can surface moments worth opening and disclose the source and scan coverage. A named reviewer records finding resolutions, and AI does not independently set the candidate's score or outcome.
No. SkillCort does not use video or audio to infer emotion or personality. Camera, screen, audio, activity, and identity signals are limited to the disclosed integrity purpose configured for the assessment.
Yes. Every active control — recording modes, secure exam mode, AI-assisted review — is named on the consent screen before the assessment starts. Capture that isn't consented doesn't run.
Every proctoring event is hash-chained per attempt as it's written; editing, deleting or reordering breaks the chain detectably. The audit pack re-verifies the chain and reports exactly where any break occurs.
The session selfie, photo ID and optional reference photo can be shown together for review. Where enabled, a one-to-one similarity result may compare the submitted face images as a signal. It is not a gallery or cross-candidate search, and a named reviewer records the identity decision.
Recordings live in isolated storage with per-workspace retention windows and per-attempt erase. High-sensitivity media isn't kept longer than the window you set.
Protect fairness without hurting trust.
Layer integrity to your stakes, connect findings to their source, and keep reviewer resolutions and candidate outcomes under human ownership.