Rubric and Readiness

Score each dimension 0 (missing/incorrect), 1 (partial, unexplained or dependent on an undisclosed solution), or 2 (correct, reproducible and independently explained). Require 2/2 on every dimension; strengths cannot average away a missing boundary. Repair the failed dimension with a fresh variant and preserve the original attempt.

AI1 — Model and context literacy

Explain model/context/tool distinctions and state evidence needed for a claimed external action.

AI2 — Assistance quality

Preserve a first attempt, focused request, checked response and fresh unassisted explanation; label live versus simulated evidence.

AI3 — Evaluation

Implement schema and behavioral evaluation, catch known bad proposals and include an independently expected extra case.

AI4 — Transfer and control

Pass the changed-threshold task, explain counterexamples and identify the authority/permission boundary for untrusted content.

Integrity and limits

A supplied checker pass is necessary for the corresponding implementation, not sufficient for the whole module. Distinguish predictions, authored fixtures, actual runs and independent explanations. Record assistance and unresolved limitations. A blocked required live check remains pending; continue other independent work without claiming completion. Repeated repairs and optional full-course work are outside the stated first-pass allowance.