TRD-06 · The non-automotive variant

Pick a trade outside the shop bay — culinary, cosmetology, health support, early childhood, horticulture — and define where AI assist is safe, where it is merely useless, and where it is prohibited. Then prove the boundary with a test rather than an opinion.

One trade and one task family the teacher approves. The task family must have a written safety authority: the FDA Food Code and the program's food-safety standard for culinary (https://www.fda.gov/food/retail-food-protection/fda-food-code), manufacturer directions for chemical services, scope-of-practice documentation for health support, or the licensing standard for early childhood. At least ten scenarios in the task family with an established correct answer, including scenarios that cross a safety or scope line.

Do not prepare food for consumption, apply a chemical service to a person, or perform a patient-care task on the basis of a model's suggestion. Instructor sign-off governs every hands-on step.

Choosing a task family with no written authority, leaving nothing to score against. Testing only safe scenarios, so the boundary never gets probed. Treating a confident answer about allergens, chemical interactions, or scope of practice as informational. Assuming an automotive framing transfers — a food-safety violation and a misdiagnosed no-start fail in different ways and on different timescales.

A boundary document: the trade, task family, and safety authority named; the scenario results, with the line-crossing cases called out individually; a three-way classification of AI assist in this trade — safe with verification, useless, prohibited — each item backed by a scenario; and the refusal list, with the harm named for each entry.

CT: distinguishing a safety boundary from a preference. CR: choosing scenarios that actually test the line. CO: a peer challenges one classification and it is defended or moved. CM: a boundary document a program lead could adopt. CZ: the person who eats, wears, or receives the outcome.

A credentialed practitioner in that trade — chef, cosmetologist, nurse, or licensed provider — reviews the classification. Standing instruction: reject any "safe" classification not backed by a tested scenario. School-supervised.

R1: one trade, one task family, one written authority. R2: scenarios and scoring reproducible. R3: comparator is the authority's own procedure. R4: line-crossing cases identified and counted. R5: document is adoptable. R6: prohibited list is specific and justified.

(a) A general "be careful with AI in healthcare" essay with no scenarios. (b) The task family has no written authority, so every classification is opinion.

Lane A immediately (culinary, cosmetology, health-science, or early-childhood capstone). No verified credit claim.