Prebuild — from an idea to a costed build plan

What this is, and what it isn't.

It does first-order arithmetic, shown in full, so you can check it rather than trust it. It is not CAD and it is not a simulator.
It does not tell you whether your idea will work. It tells you what it would take, and lets the number speak.
Every figure is labelled measured observed judgement or yours. Where a real price or spec isn't known, you get a placeholder marked as an estimate — never an invented number.
It sells nothing, links to no shop, and takes no affiliate fee.
Its whole purpose is to get you to the bench sooner, with fewer surprises waiting there.

Stage 1 — Define the thing

Almost every hardware project that dies had one governing number nobody wrote down. This stage is about finding yours, and then — the part people skip — working out what actually sets it.

The object

What is it? What must it do? One sentence, in the language you'd use to a friend. If you can't finish this sentence, that's the finding — go and finish it before anything else.

The governing number

The single quantity that decides whether this works at all. Everything downstream is arithmetic hanging off it, so if it's wrong, everything is wrong — and it will still look fine.

What kind of number is it? Your value Unit

Where did that number come from? Be honest here. “It felt about right” is a legitimate answer and a useful one — it tells you which number to go and check first.

What sets that number?

This is the question the worked example exists to teach. Makers almost always assume the governing number is set by the headline component — the motor, the chiller, the battery. It's usually set by something duller and further upstream, and that something is usually an interface: how much of the thing can actually accept or deliver what you're pushing through it.

Which of these do you think is the real limiter?

Stage 2 — Derive the constraints

From the governing number, the physics that follow whether you like them or not. This is where the tool earns its existence: it does the arithmetic a first-timer skips, and shows the working so you can check it rather than trust it.

measured printed on a data sheet or in a peer-reviewed paper

observed read off a graph, or inferred from product naming

judgement reasoned, not sourced

yours a number you typed

The fork

There is no single derivation, because there is no single architecture. Pick a branch and everything below recomputes. The branches interact — a choice that saves mass in one subsystem often spends it in another — so the comparison table underneath shows every combination rather than only the one you are looking at.

Cold engine — how the heat gets moved Battery — chemistry and how it's built Run time wanted (h) The single input that decides which branch wins.

Inputs

Defaults are the sourced figures from the worked example. Change any of them and every number below moves. Where a default came from a real data sheet, the source is listed at the bottom of this stage.

Cooling delivered, Qc (W) Pump draw (W) Electronics draw (W)

Buffer tank volume (L) Tank temperature rise (K) Solar area (m²) Panel output at STC (W/m²)

Where the default numbers came from — check them yourself

Cooling vest heat flux — Ciuha, Valenčič & Mekjavić (2020), Cooling efficiency of vests with different cooling concepts over 8-hour trials, Ergonomics 64(5) 625–639. Thermal manikin, 35 °C, 35% RH, torso contact area 0.54 m². Best sustained cooling power of any vest tested: 41 W/m² (air-cooled). Best liquid-cooled vest: 17 W/m² average, 60 W/m² peak. Retrieved via PubMed. doi:10.1080/00140139.2020.1853820
Full-body liquid cooling garment — NASA Liquid Cooling and Ventilation Garment covers 1.776 m². Apollo 12 EVA removed 894.4 Btu/h (262.1 W), which is 148 W/m². encyclopedia.pub/entry/28482 · NASA NTRS 20160005252
Chiller module — Aspen Systems LCM-600: 2.7 kg, 191×161×134 mm, 360 W max at 24 VDC, 15 A, R-134a. LCM-600 data sheet · LCM-650: 2.9 kg, 2 L/min water, 150 cfm airflow, data sheet · FP00116 Q9-LCM: 2.85 kg, up to 17.0 A at 24 VDC, requires a customer-supplied coolant pump, manual
Cells — Samsung SDI INR21700-50S: 5.0 Ah, 3.6 V, 72.0 g max → 250 Wh/kg. specification · Molicel INR-21700-P45B: 4.5 Ah, 3.6 V, 70.0 g max → 231 Wh/kg. specification
Solar — SunPower SPR-E-Flex-110: 110 W, 1165×556 mm (0.648 m²), 2.0 kg → 170 W/m² at STC, 3.09 kg/m². data sheet
Pump — TOPSFLO TL-B10, 12/24 V, 7–12 L/min, 3–8 m head, liquid to 100 °C. Mass and wattage are not printed on the page, so both are judgement here. product page
Metabolic rate — rest 115 W, light 180 W, moderate 300 W, heavy 415 W, very heavy 520 W, for a 70 kg worker. Oregon OSHA heat illness appendix
Specific heat of water — 4186 J/(kg·K).

Stage 3 — Build the parts list

Subsystem by subsystem, with a role for every line. Prices you haven't checked stay marked as estimates. And the padding row stays, because every hardware post-mortem says the same thing: the cost that kills you is the one you didn't list.

Subsystem	Part	What it's for	Qty	Unit	Line	Price is

Padding (%) Fasteners, connectors, wire, adhesive, the third one you break.

Stage 4 — Cost the run

A roll-up of your stage 3 parts list across five run sizes. It invents no supplier discounts. Your entered prices carry through unchanged at every volume, because a discount you have not been quoted is a wish. What changes with quantity here is only the arithmetic you can honestly do at a desk: tooling spread thinner, and the per-unit costs that have nothing to do with parts.

The costs that are not parts

Tooling, total ($)Moulds, jigs, fixtures. Paid once, spread across the run. Freight per unit ($) Duty / tariff (%)

Fulfilment per unit ($)Pick, pack, box, postage to the backer. Platform + payment fees (%)Kickstarter is 5% plus payment processing. Failure / rework allowance (%)Units you build twice.

Stage 5 — Plan the iterations

Four versions, each with one question it exists to answer and one exit test — the thing that has to be true before you are allowed to move on. Write them before any more arithmetic.

Why the gate matters more than the plan. Hardware projects run long because a version gets declared finished on a feeling rather than a test. A gate written in advance, by you, in a calm moment, is the only defence against your own optimism at 2am three weeks from now.

The gate people delete is v1.5. Surviving a demo is a different engineering problem from working — it means working repeatedly, in a room you do not control, with someone watching, after being carried there. Skipping it is why prototypes fail on camera.

Load a saved JSON A working paper you can take to a supplier, a teacher, or a camera. Not a certificate.

Licence. MIT. Fork it, teach with it, strip out the worked example and put your own in.

Privacy. This is one HTML file with no network requests. There is no server, no account, no analytics and no storage — nothing typed here can leave your device, because there is no code in this file that could send it. Don't take that on trust: open the source and search for fetch(, XMLHttpRequest, sendBeacon, <script src, <img, localStorage. The links out are the only network traffic, and they only happen if you click one.

A caution worth stating. Everything here is first-order arithmetic. It gets you to the right order of magnitude and shows you which assumption is load-bearing. It is not a substitute for a real engineer, real component testing, or a real bench. Where a number matters — a safety margin, a pressure, a current, anything that could hurt someone — go and measure it.

Intended for young adults building their first piece of hardware, with a few weekends and a few hundred dollars of runway. The whole job of this tool is to help you spend that runway on the right thing.