Integrity Layer
Integrity in layers. Judgment by people.
From a browser-native secure exam mode to human-adjudicated AI review of recordings, SkillCort layers integrity to the stakes of each assessment. AI and signals flag moments worth looking at; a person reviews every flag and makes every call.
AI flags. Humans decide. A single signal never fails anyone.
Four layers, each sized to the stakes
Lockdown in the browser — nothing to install.
A secure exam mode inside the normal browser: fullscreen is enforced with your choice of policy (warn, pause the exam, flag for review, or auto-submit at a violation limit), and the session is hardened against the usual escape hatches.
- Fullscreen enforcement with per-assessment policy
- Keyboard lock (Esc and system shortcuts included), print & PrintScreen detection
- Second-display block at entry and mid-exam detection
- Stopping screen share locks the exam until sharing resumes
- Virtual-machine suspicion, DevTools attempts, copy/paste blocking
- A session watermark (candidate + timestamp) deters leaked screenshots
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, the constant is the same: signals and AI add context for a human reviewer. They never act as an automatic verdict.
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.
Human verdict
A person confirms or dismisses every flag, on the record.
The stack at a glance
Secure Exam Mode
Browser-native lockdown: fullscreen policy, keyboard lock, print & DevTools guards.
Live Monitor Wall
Watch active sessions live — webcam and screen frames as they arrive.
Honeytokens
Invisible markers catch question text that round-trips through an external AI.
Hash-Chained Event Log
Every proctoring event is chained; tampering is detectable and provable.
Collusion Analysis
Identical wrong answers and near-duplicate essays surface across the cohort.
Human-Adjudicated AI Review
AI flags webcam, screen and audio moments; a person rules on each one.
Proportional Levels
From light signals to the full stack — set per assessment, disclosed to candidates.
Audit Pack
Integrity grade, chain verification and the full timeline, exportable per candidate.
Reports & exportsLockdown alone is not integrity.
Lockdown-only tools
- One aggressive setting for every assessment
- Signals become automatic failures
- No evidence trail a candidate could contest
- AI verdicts with no human in the loop
SkillCort Integrity Layer
- Full lockdown when stakes demand it — proportional layers when they don't
- Deterministic forensics: honeytokens, hash-chained logs, collusion analysis
- Every AI flag gets an explicit human verdict, on the record
- 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
- Identity verification is a one-to-one comparison against the candidate’s own reference photo — no galleries, no stored face templates, and a person always makes the identity decision
- No cross-client candidate memory or hidden reputation scores
- No automatic disqualification, ever
- Full audit trail for every flag and every verdict
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.
Never. AI review runs in the reviewer's context, flags moments worth opening, and every flag receives an explicit human confirm-or-dismiss verdict. Runs are provenance-logged, coverage is disclosed, and the whole pipeline is production-verified and covered by our public governance report.
No. SkillCort does not perform covert behavioral profiling or emotion and personality inference from video or audio. Optional identity verification compares a snapshot only against the candidate’s own reference photo (one-to-one) — there is no matching against galleries or other people, no biometric template is stored, and a human reviewer always makes the identity decision. These boundaries are published in our AI governance statement.
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.
Flags route to human review — never an automatic rejection. A structured in-product candidate appeal flow is in progress; today, candidates raise concerns through the hiring team's contact channel.
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 — and keep every verdict human, on the record, and checkable.