TRD-07 · Parts and supply chain

Photograph a part, or hand over a spec sheet, and ask a model to identify it. Measure the mis-identification rate — then price what a wrong part actually costs in return shipping, delay, comeback, and a customer waiting on a vehicle or a building.

Thirty or more parts with verified identity — school-shop inventory, a supplier catalogue, or instructor-supplied components. Photographs taken under the conditions a technician actually works in: shop lighting, mounted position, dirt, partial obstruction, worn or missing labels. A clean studio photograph measures nothing about a real parts counter. An AI tool the school permits, named with version.

Photographing only clean, unmounted, well-labeled parts. Scoring a near-miss as correct when the part is dimensionally or electrically incompatible. Counting the part price as the cost while ignoring the labor, delay, and second trip. Treating a confident identification as verified when the label was never visible.

An identification study: all parts with verified identity and photo conditions recorded; the model's identification scored exact / compatible / incompatible, since near-misses are the interesting result; the mis-ID rate broken out by condition (clean versus dirty, labeled versus unlabeled, mounted versus loose); a costed example of one wrong-part order carried through to the customer; and the verification step that stays mandatory regardless of the tool's confidence.

CT: distinguishing a plausible identification from a verified one. CR: photographing under real conditions. CO: a peer verifies ten identifications against the catalogue. CM: a parts-counter rule stated in one line. CZ: the customer whose vehicle sits an extra three days.

A parts manager, counter professional, or shop foreman reviews the scoring categories and the cost example. Standing instruction: reject a study built on studio photographs. School-supervised.

R1: one part family, thirty-plus items, conditions recorded. R2: identities verified against a catalogue. R3: comparator is catalogue lookup by number. R4: near-misses counted; downstream cost included. R5: the counter rule is one line. R6: names what is never ordered on a model's say-so.

(a) Clean photographs, a 90% rate, and no relationship to a real parts counter. (b) Only the part price is counted, so a $12 error looks cheap when it cost three days.

Lane A immediately (center capstone or a logistics or business cross-listing). No verified credit claim.