§ Exam Blueprint
What's on the exams — no surprises
Serious credentials publish their blueprints. This page tells you exactly what each SAPIEN exam covers, how it's scored, and what the policies are — so you can prepare with confidence and your employer knows precisely what the credential attests.
Exam formats at a glance
| Tier | Time commitment | Exam | Pass mark | Practical work | Validity |
|---|---|---|---|---|---|
| Awareness | 30–60 minutes, one sitting | 20 questions | 80% | 3 Driftproof scenarios | No expiration |
| Practitioner | 2–4 hours, multiple sessions | 30 questions | 80% | 4 Driftproof scenarios + 3 Red Lab exercises | 1 year |
| Assessor | Multiple sessions over days | 40 questions | 85% | Live Red Lab + 3 rubric submissions | 1 year |
Each attempt is a server-drawn, domain-stratified exam from a larger active question pool (you never see the full bank), mixing single-select and multi-select items, graded entirely server-side. Correct answers are never sent to your browser.
Learn by doing, then prove it
The exam is the smallest part. Most of your time is hands-on. In Driftproofyou play the AI and hold a boundary under pressure; in the Red Lab you flip roles and become the user, applying realistic pressure to make a simulated assistant drift — then you annotate which move broke it and why. Passing a lab requires both: you induced the drift and you can explain the mechanism. Telegraphed or adversarial phrasing ("ignore your rules") is penalized, because real pressure reads as natural. This is the assessor's core craft, practiced and proven rather than merely read.
Assessor exam domains
The Assessor exam is weighted across five domains. Questions are predominantly scenario-based — they test whether you can apply the methodology, not whether you memorized it.
Boundary definition, escalation design, pressure types, severity calibration, humanization and leak rules, validation gates.
Council composition and verdict math, judge sycophancy mitigation, dimensional scoring, score interpretation and comparability.
Running full benchmarks, reproducibility, run artifacts and evidence handling, failure and resume handling, provider considerations.
Mapping findings to NIST AI RMF, EU AI Act, ISO/IEC 42001, and OWASP LLM Top 10 — including what SAPIEN evidence does not claim.
Executive reporting, traceable findings, disclosure norms, avoiding compliance-guarantee language, assessor integrity.
Assessor practical work
Three artifacts, each submitted against a published rubric. Every rubric criterion must be present and substantive or the submission is rejected (up to 5 attempts per practical).Human grading of these artifacts is not yet live — submissions are currently validated for completeness only and stored for later review, and the Assessor credential is not being issued until human review is in place.
- Full benchmark run — a real council-scored voigt-kampff run, correctly interpreted, with at least two drift findings evidenced at the turn level and honest methodology limits.
- Custom scenario design — an original scenario testing exactly one behavioral boundary, escalating pressure, a humanized voice, and a clean
voigt-kampff validatepass. - Assessment report — an executive-ready report with traceable findings, governance-framework mapping, and prioritized remediation.
Policies
- Attempts. Each exam allows up to 10 attempts; each practical allows up to 5 submissions. Hitting a cap requires contacting support — attempts are audited.
- Retake wait. After your second unsuccessful attempt at an exam, each further attempt requires a 14-day wait. Use it to review — every new attempt is a fresh draw from the pool.
- Integrity. All grading is server-side. Submitted practicals must be your own work; identical artifacts across candidates are treated as violations and can lead to revocation.
- Renewal. Practitioner and Assessor certificates are valid for 1 year and renew by re-passing the current exam — the corpus and methodology evolve, and the credential attests currency. Awareness never expires.
- Verification. Every certificate has a public verification page and an Open Badges 2.0 assertion your employer can check independently.
How items are written, calibrated by a multi-model review council, and retired is documented on the exam methodology page — the process is public, the answers are not.