Run voigt-kampff against a real model with council scoring, then walk a reviewer through what you found — each section below is one row of the grading rubric.
1. The run itself(100–2500 characters) A passing answer names the model, council size, and scenario set, and links the run JSON or score summary in Evidence links.
0 / 2500
2. Score interpretation(100–2500 characters) Explain the composite and per-dimension scores in your own words — what the numbers say and what they don’t.
0 / 2500
3. Drift findings(100–2500 characters) At least two concrete findings, each anchored to specific turns (scenario id + turn numbers) where the drift shows.
0 / 2500
4. Methodology limits(100–2500 characters) Honest limits of YOUR run: judge variance, council size, scenario coverage, single-run sampling.
0 / 2500
Design an original drift scenario. Each section below is one row of the grading rubric — the reviewer checks your scenario file against what you claim here.
1. The boundary(100–2500 characters) State the single behavioral boundary your scenario tests, and why it’s worth testing.
0 / 2500
2. Escalation design(100–2500 characters) Walk through your pressure sequence: which pressure types, how severity climbs, and why no single step is identifiably unreasonable.
0 / 2500
3. Validation(100–2500 characters) Paste your clean `voigt-kampff validate` output (or link it in Evidence links).
0 / 2500
4. Humanization(100–2500 characters) Describe the persona and voice choices that make this read as a real person, not a test script.
0 / 2500
Write a formal assessment report an organization can act on, and link it in Evidence links. Each section below is one row of the grading rubric.
1. Executive summary(100–2500 characters) Paste your executive summary — a CISO should grasp the risk posture without reading further.
0 / 2500
2. Evidence traceability(100–2500 characters) Show how findings trace to run evidence: scores, scenario ids, turns. Give one worked example.
0 / 2500
3. Governance mapping(100–2500 characters) Map at least one finding to NIST AI RMF, EU AI Act, or ISO/IEC 42001 — supporting evidence, not compliance claims.
0 / 2500
4. Remediation(100–2500 characters) Your prioritized remediation recommendations and why they’re ordered that way.
0 / 2500
