adulbrich
From the four-skills audit (critical thinking, AI literacy, leadership, collaboration). This is the highest-leverage item in it: the content already exists on every page, and this converts AI literacy from taught to assessed without restructuring anything. ## The finding - **10 of 10** assignment pages pose an AURA claim for the student to judge. - **1 of 10** rubrics mentions AURA at all (`household-energy-panel`, in its reflection criterion). - Even there it is completion-graded. The assignments skill states the rule: "paper practice and reflection on completion." So the course teaches students to distrust AI output and then never grades whether they can. The one skill most specific to this revision of the course is the one skill with no correctness criterion anywhere. ## Scope - [ ] Add a criterion to each of the ten rubric TSVs in `canvas/assignments/` that scores the AURA judgment on **correctness**: did the student identify what is actually wrong with the claim, in terms of the computation or the method, rather than merely asserting AURA is wrong? - [ ] Give it a partial level. "Spotted that it is wrong but not why" is a real intermediate state and is worth distinguishing from both ends. - [ ] Update `.claude/skills/engr103-assignments/SKILL.md`, which currently sends the AURA sentence to a completion-graded reflection, so future assignments inherit the criterion. - [ ] Rebalance the point split so the total still lands near 100; the implementation criterion at about 5 points is the natural donor. Note the TA cost: this is a read-and-judge criterion, not an autograded one. It is the same kind of work as the existing implementation criterion, which a TA already reads. --- Audit set: #29 grade the AURA judgment, #30 equity clause, #31 peer review, #32 activity pairing protocol, #33 recommendation memo. #34 records the three changes that would reverse a VISION decision.