# Day 3 Lab — Representative Risk-Tiering Output
# Captured from a correct run of: docker compose run --rm runner python app/risk_tiering.py
# Pure Python, no model, fully offline (no network calls ever).
#
# The classification is DETERMINISTIC — the same scenarios.json always produces
# the same tiers. Key assertions:
#   - S1, S2 => prohibited   (Art. 5 practices)
#   - S3, S4, S5 => high-risk (Annex III / safety component)
#   - S6, S7 => limited      (Art. 50 transparency)
#   - S8, S9, S10 => minimal (no Annex III, no transparency trigger)

=== Day 3 Lab — AI Risk Tiering: NIST AI RMF + EU AI Act ===

Loaded 10 scenario(s) from data/scenarios.json.

EU AI Act risk tiers
ID   System                                   Sector                    EU tier
S1   Government citizen social-scoring plat…  public administration     prohibited
S2   Real-time public facial recognition …    law enforcement           prohibited
S3   AI resume-screening assistant (HR)        employment                high-risk
S4   AI credit-scoring for consumer loans      financial services        high-risk
S5   AI managing access to power-grid SCADA    critical infrastructure   high-risk
S6   Customer-facing security help chatbot     security tooling          limited
S7   Marketing synthetic-media generator       marketing                 limited
S8   Internal phishing-email classifier (SOC)  security tooling          minimal
S9   Internal AI meeting-notes summarizer      productivity              minimal
S10  Autonomous incident-response host isol…   security tooling          minimal

Tier distribution: prohibited=2  high-risk=3  limited=2  minimal=3

[Per-scenario Assessment panels + NIST AI RMF (Govern/Map/Measure/Manage) tables follow]

Wrote 10 filled worksheet(s) to /workspace/worksheets/worksheet_<id>.md

# Note on S10 (autonomous host isolator): tiers as MINIMAL under the EU AI Act
# (it is an internal security tool, not an Annex III use case) — but its autonomy
# still triggers extra NIST MANAGE actions (human-in-the-loop override, kill-switch).
# Tier and control-depth are related but not identical: a minimal-tier system can
# still warrant strong operational controls.
