Rubric and Readiness
Score each dimension 0 (missing or incorrect), 1 (partial, unexplained or dependent on an undisclosed solution), or 2 (complete, observable and independently explained). Require every dimension at 2. Preserve failed and repaired work; confidence and speed are not grading criteria.
AL1 — Model and task choice
Explain response versus evidence versus action, and write a bounded tutoring request with an explicit check and authority limit.
AL2 — Verification and revision
Independently check the fresh plan’s totals and constraints, identify unsupported claims, and preserve an evidence-based correction.
AL3 — Learning and human control
Show an initial attempt and independent changed-case explanation, distinguish assistance sources, minimize private data and reject instruction-like content as authority.
Calibration limits
Use answer-key guidance to distinguish incomplete evidence from a defensible result. A plan is not an observation; a fixture is not a live system. A blocked practical gate remains pending. One bounded repair is included in the allowance; further repairs may extend it.