Concept study · build execution software
A working prototype of the workflow that internal manufacturing software lives or dies on: a technician executing one step of a build against a hardware kit, and the planner who has to know the second that step goes sideways. Built as an argument, not a mockup — it runs, it measures itself, and the two surfaces talk to each other.
Pass 1 — snug
22 in-lbf
Pulling position 1
The part the bolt pattern sits on — eight fasteners on the bolt circle, the clevis lug and pin bore, seated against the panel rail.
NAS1149D0463K
Wrong part in bin K-2291-B
Stack-up
4 of 7The part the bolt pattern sits on — eight fasteners on the bolt circle, the clevis lug and pin bore, seated against the panel rail.
Live components, not screenshots — the same code the technician surface renders, re-resolving its tokens against the floor palette inside this page.
Five steps of a hinge-bracket install. Photograph the kit tray, get told two lines are wrong before you start, page a runner, torque in enforced sequence, sign off.
Start the traveler Desk · cursor · many rowsWIP by station, what is over plan, the MRB queue, and the shortage the technician just raised — arriving live, in the same tokens at four times the density.
Open the boardBest viewed as two windows side by side.
Put the technician view in a narrow window and the planner board beside it. Run the kit check and watch the shortage appear on the board while you are still standing at the bench. That loop — floor to supply chain and back — is the thing worth arguing about.
Want the reasoning, not just the screens?
Research approach, the six decisions in full, what I'd measure, and what this prototype does not prove — on automaticdelight.com.
The rationale is the deliverable. The screens are evidence.
Green, amber and red mean go, hold and stop — and nothing else. No brand accent, no decorative highlight, no chart series may borrow them. The moment a color means two things on a factory floor, it means nothing.
Raising a non-conformance is one tap with a reason code and optional prose. Signing a buy-off into the permanent build record is a deliberate 1.2-second hold. If reporting a problem is harder than ignoring one, you get clean data and bad hardware.
Part numbers and serials get read aloud, transcribed onto a card, and typed into a different system. NAS1149D0463K versus NAS1149D0332K is a scrap event. Proportional type makes that comparison harder for no reason.
One step per screen, fixed status rail at the top, fixed commit rail at the bottom. A technician at a mounted tablet should never scroll to find out how to proceed, and should never lose their place to report something.
The traveler has always printed the torque sequence. Enforcing it is different work: a wrong pull is caught in the second it happens instead of during a QA re-walk two hours later, when the fix is rework instead of a tap.
The technician and planner screens share every token and component. Only the density and contrast variables change. That is the whole mechanism for staying coherent as modules multiply — not a rule in a doc, a variable in a file.
Everything here is drawn from one token file. There is no arbitrary type size and no raw hex anywhere in a component — and that is enforced, not requested: a precommit guard reads the staged diff and rejects new arbitrary values, while a ratchet lets existing debt pay itself down instead of blocking a shipping team. A design system that only lives in a library file is a suggestion. This one fails the commit.
$ npm run guard
components/StepView.tsx
✕ arbitrary font-size — use a named text-* token
✕ raw hex — use a color token
2 new violations. Commit blocked.An unsolicited concept study by Bryan Holland, built to think through shop-floor build execution properly rather than describe it in a deck. It is not affiliated with, endorsed by, or built from the internal software of any company, and contains no proprietary information — every part number, torque value and work order in it is invented. The measured comparison on the summary screen uses an illustrative baseline for the paper-traveler flow; the real number would come from your floor, which is exactly the point of instrumenting it.
Next 16 · React 19 · Tailwind v4 · no component library · roughly a day of work.