AI-Assisted Trades & Credentials · Capstone brief
TRD-10 · Shop data privacy
Customer names, addresses, plate and VIN numbers, building layouts, security details, household schedules, patient information. Write the rule for what never enters a public model — and make it something a busy technician will actually follow at 4:45 on a Friday.
The question
Customer names, addresses, plate and VIN numbers, building layouts, security details, household schedules, patient information. Write the rule for what never enters a public model — and make it something a busy technician will actually follow at 4:45 on a Friday.
System / materials
Your program's actual data: the fields a repair order, service ticket, intake form, or client record contains. The division's data policy and any placement agreement the program operates under. This brief writes a policy; no real customer data is used as an example, including in the appendix.
Expected failure modes
Producing a policy that just says "do not share customer data," which fails at the first real decision. Ignoring the field combinations that identify someone even when no single field does — a partial address plus a vehicle description plus a service date. Missing the informal channels where it actually leaks: a phone photo of a work order, a screenshot pasted into a chat, a personal account used because the shop tool was slow. Writing a rule with no permitted path, which guarantees it gets bypassed.
Done looks like
A shop data rule: a field-by-field classification of everything on the shop's forms — never, only-with-approval, or fine; the identifying-combination cases named explicitly; the redaction procedure a technician runs before pasting anything, short enough to be followed; the permitted path — what to use instead when a model would genuinely help; and the leak-channel checklist covering photos, chat, personal devices, and personal accounts.
Five C's
CT: recognizing identification by combination. CR: designing a rule fast enough to survive a busy Friday. CO: a peer tries to route around the rule and the gaps get closed. CM: written for a technician, not a compliance officer. CZ: the customer whose address and absence schedule ended up in a public tool.
Mentor role
A shop owner, service manager, or the division's data or compliance lead reviews the rule against real agreements. Standing instruction: reject any rule with no permitted path — a policy that only forbids gets ignored. School-supervised.
Rubric calibration
R1: one program, one form set, one rule. R2: another student could apply the classification unaided. R3: comparator is current practice, with gaps named. R4: combination cases and informal channels both covered. R5: the redaction procedure fits on a card. R6: names what never enters a model under any circumstance.
Two ways this goes wrong
(a) A one-line prohibition that answers no real question. (b) Field-by-field rules with no attention to combinations, so the policy is satisfied and the customer is still identifiable.
Credit lane fit
Lane A immediately, and worth assigning early in any cohort running the rest of this bank. No verified credit claim.