Why isn't the prototype code the handoff?

Because production-shaped code carries every choice the model made, in a form engineering reads as decided. It runs and it looks like the product, so nobody asks what it does on the bad path. Reddit's feature capsule ends in a draft PR that, in Blair Metcalf's words, may or may not go through.

Should prototype code be kept or rebuilt?

Per layer. Keep what came from the design system mapping, since it was production code before the prototype began. Rebuild the logic, the data handling, and whatever the prompt decides. Uditi Sharma says the answer probably depends on the product's risk, and the layer is where the risk sits.

How big is the test package?

The sizes I use are three to five rubric criteria, at least twenty test cases, graded binary per dimension, plus one hard-fail row per never-do line. Thirty minutes gets the first five never-do rows written.

What does engineering owe back?

A harness that runs design's rows on every change, a regression report that names the failing row and shows the output, and the list of rows it cannot pass inside the latency or cost budget, returned to design and the PM as a decision.

Who breaks the tie when a design row and a cost budget collide?

The PM. Design owns the row and its threshold, engineering owns the budget, both names go on the row, and the PM decides because the PM owns the outcome the trade-off is about.

What Design Hands Engineering Instead of a Mockup

What should design hand to engineering instead of a mockup?

THE SHORT ANSWER

The behavior spec, rewritten as tests engineering can run. Never-do lines become hard-fail rows where one failure blocks the merge, the acceptable range becomes graded rows with a real example at the floor, the wrong path becomes fixtures for refusal, low confidence, and recovery, and the rubric becomes the grader. The prototype goes along as a reference, never as the source, and its code is kept or rebuilt per layer.

A mockup was honest about being a picture. Production-shaped code isn't. It runs, it uses the real components, and it carries every choice the model made, so engineering treats it as decided.

My rule, from part 3 of The Design Transition, Reddit's Handoff Ships a Draft PR. Design Should Hand Engineering the Tests.: hand over the behavior spec as rows.

Never-do lines become hard-fail rows. One failure blocks the merge.
The acceptable range becomes graded rows with a real example at the floor.
The wrong path becomes fixtures: refusal, low confidence, timeout, recovery.
The rubric becomes the grader.

Add the ratified decision ledger from part 2, so engineering knows which choices someone made on purpose.

Engineering owes back a harness that runs the rows on every change, a regression report that names the row, and the rows it can't pass inside budget, as a decision for design and the PM. Nielsen Norman Group's line from the same week is why it matters: a task made faster for one person can create work for another.

Every part of the series is at the series hub.

Last reviewed 2026-10-28 · 1 min read

What should design hand to engineering instead of a mockup?

SOURCES

THE LONG VERSION

RELATED ANSWERS