Guides / Design and Fit

The Prototype That Saves a Group Buy

September 21, 2026

A prototype costs ten to twenty times what a production unit costs, and it is still the cheapest thing in the project. The reason is arithmetic: finding a fitment error on one part costs you a few weeks and a few hundred. Finding it on a hundred and eighty parts costs you the run.

What a prototype is for

Not for checking that the shop can machine aluminium. It can.

A prototype answers questions that no model can answer, because they involve real materials, real coatings and real assembly:

• Does the plate drop into the pocket after both surfaces are coated?
• Does the gasket compress the way you calculated, and does the board sound like you intended?
• Does a thick USB-C cable fit the cutout, not just a thin one?
• Do the case halves close with an even seam all the way round?
• Is the colour the colour?
• Does the weight sit flush in its recess, or is there a visible gap on one side?

Every one of those is a real failure that has ended real group buys. None of them show up in CAD.

Insist that it is made the production way

A prototype that is machined but not anodised tells you almost nothing about fit, because coating thickness is exactly what catches people. A prototype cut on a different machine with a different fixture tells you less than you think about the production parts.

Ask for the sample to be made the way production will be made: same alloy, same setups where possible, same blast, same coating, same colour. It costs a little more and it is the entire point.

Assemble it properly, with your own parts

Do not inspect a prototype on a desk. Build it.

Put your actual PCB in it, screw down the plate, fit the gaskets, install switches and keycaps and type on it for an evening. Then take it apart and put it back together, because reassembly finds things that assembly does not.

Look at it in daylight and under a desk lamp, because anodised colour changes character between the two and your buyers will see both.

Write down what is wrong, once

Make one list. Every issue, with a photograph and a measurement where you have one. Send the whole list at once rather than five emails over a week.

Separate it into three groups:

1. Must change before production. Fitment, function, anything that would generate returns.
2. Should change if it is cheap. Cosmetic details that bother you.
3. Noted, not changing. Things you have decided to live with.

That third category matters more than it looks. It stops the same conversation happening three times, and it tells the shop you are making decisions rather than accumulating complaints.

Decide whether you need a second round

A second prototype after changes is common and often correct, particularly if the first one revealed a dimensional problem rather than a cosmetic one.

Budget for two rounds in the schedule and be pleased if you only need one. Adding a second round to a timeline that assumed one is where public ship dates start slipping. There is more on schedule realism in how long a run actually takes.

Keep the approved sample

Once you sign off a prototype, that physical object becomes the standard. It is the golden sample: the definition of the colour, the texture and what an acceptable part looks like.

Both sides should have one where possible. Keep yours somewhere safe and unscratched. When a question comes up in month four about whether a batch is too light, the answer is the sample, not the email thread. How to write that into an agreement is covered in agreeing what counts as a defect.

The cost of skipping it

Say you skip the prototype on a run of 180 cases at 90 each. Production finds that the plate pocket is 0.15 mm too tight after e-coating.

The cases are already machined, anodised and packed. There is no rework on a coated cosmetic part, so the options are stripping and recoating at the shop’s mercy, machining the plates smaller and hoping the gap does not show, or remaking. Any of those is weeks, and the first two are compromises you will be explaining publicly.

The prototype that would have caught it cost around a thousand and three weeks. That is the trade, and it is not close.

Frequently asked questions

Keep reading

• Machining a Knob for a Keyboard Encoder, and the Case Hole It Sits In
• Sizing a USB-C Cutout for the Cable Boot and the Daughterboard Behind It