Prebuild — Divergent Futures

Work out whether a hardware idea is buildable, what it takes and what it costs — before you spend your one first impression. Finds the one number the whole thing hangs on, then shows you what it actually costs.

Almost every hardware project that dies had one governing number nobody wrote down. Not a hard number — an unexamined one. Everything downstream is arithmetic performed correctly on a wrong input, which is why it stays plausible right up until it doesn’t.

Prebuild makes you name that number, then checks it against the thing that actually sets it. That check usually contradicts the number, and the contradiction is the point.

What it does

Finds the governing number, then finds what really governs it. In the worked example the claim was 150 W of cooling, reasoned from how much heat a body produces. Nobody checked how much a garment can accept — which turns out to be about 25 W. The number wasn’t wrong by a bit. It was wrong by 6×, and it was wrong upstream of every calculation that followed.

Derives the constraints with the working shown. Every step prints its formula and its inputs, so you check the arithmetic rather than trusting it.

Sweeps every architecture, not the one you arrived with. All 36 combinations of cold engine and energy source get a row and a stated verdict — including the ones that lose badly. An option you never listed looks exactly like an option that lost, and neither argues its case.

Remembers why you ruled things out. The ruled-out register stores your reason and the number you gave it. When that number moves, it tells you the decision deserves another look. Nothing else in a design will remind you — a rejected option never asks to be reconsidered.

Costs the run at 1, 10, 50, 250 and 1000 — with tooling amortisation, freight, duties, fulfilment, fees and rework. And it refuses to model volume price declines, because a discount you have not been quoted is a wish. The table is an honest ceiling until you replace estimates with quotes, and it tells you which three to chase first.

Makes you write four exit tests before going further: does the physics work, does it work on a person, does it survive a demo, can it be made 250 times.

The part that connects to LaunchPath

Prebuild’s most honest output — the true cost of a real run — usually lands in the funding-goal band that LaunchPath scores worst. That is not a fault in either tool. The honest number and the fundable number are usually different numbers, and the gap between them is the project. Prebuild names the gap and gives four ways to close it rather than pretending it isn’t there.

What it isn’t

Not CAD, not a simulator, and not a verdict machine — there is no “don’t build this” anywhere in it. It does first-order arithmetic, shown in full, and lets the number speak.

Nothing in it has been built or measured. It is paper end to end, and it says so.

Free, MIT licensed, and one HTML file with zero network requests — no server, no account, no analytics, no storage. Nothing you type can leave your device, because there is no code in it that could send anything. Don’t trust that: open the source and search for fetch(. The only two hits are the sentences telling you to look.