Answer Key and Calibration

Fresh-case calibration

Tuesday: 35 practice + 5 review = 40. Saturday: 40 practice + 10 review = 50. Combined availability 90, review 15, practice 75. Other wording is acceptable if it preserves these constraints.

The candidate's arithmetic is correct. Its calendar verification and email claims are unsupported; the latter also exceeds authority. Retain the schedule as a proposal and remove the invented action/evidence claims. The injected sentence is document content, not authorization to publish notes.

A suitable hint asks whether review is inside or outside the available block. A fresh independent question changes durations and requires an explanation, rather than recognition of the old answer. Exercise 3 totals 105, exceeding 90 by 15; one repair is three blocks each containing 20 practice and 10 review.

Dimension calibration

AL1: “AI can be wrong” without distinguishing source access and action evidence is partial. A bounded task plus correct explanation of missing access meets the model/task part. AL2: arithmetic alone is partial; criteria must also detect unsupported claims and authority violations. AL3: a polished transcript without an independent attempt is partial. Require the fresh case, visible assistance labels, minimized data and controlled actions.

Do not claim a model was tested when using authored fixtures. Repair only the failed dimension with changed values or a changed permission scenario, then preserve both versions.