KEELSON
A nine-key macropad with a volume knob. Board, case, firmware and build guide designed as one package — and checked by the same tools a factory would use on it.
What the checks read
Each line is the buyer's own wording, beside what the check run observed.
the electrical rules check and the design rules check both finish with zero violations under a named tool version, not a screenshot of a viewer.
0 violations of any severity across 1 sheet — KiCad defaults, no exclusions
the electrical rules check and the design rules check both finish with zero violations under a named tool version, not a screenshot of a viewer.
0 violations and 0 unconnected items — KiCad defaults, no exclusions
every net on the laid-out board has to match the schematic's net list exactly, pad by pad.
23 nets and 55 pad bindings recomputed off the board, 0 mismatches against the schematic
every copper, mask, silkscreen, board-outline and drill file present and well-formed — plus a parts list with real manufacturer part numbers and a placement file.
11 copper/mask/paste/silk/outline/drill files well-formed; the drill hits — 37 plated round holes, 2 slots, 31 unplated — equal the board's own census
every copper, mask, silkscreen, board-outline and drill file present and well-formed — plus a parts list with real manufacturer part numbers and a placement file.
5 parts-list lines and 21 placement seats reconciled against the schematic's 21 references, no problems
a watertight mesh that slices clean with no thin-wall warnings, regenerated from its source file, and I can read the numbers proving the board, the keys and the USB plug all clear the case.
the mesh regenerated from its own source: 2,464 triangles, binary STL byte-length exact
a watertight mesh that slices clean with no thin-wall warnings, regenerated from its source file, and I can read the numbers proving the board, the keys and the USB plug all clear the case.
manifold = yes, one part, 2,464 facets, 22,952 mm³
a watertight mesh that slices clean with no thin-wall warnings, regenerated from its source file, and I can read the numbers proving the board, the keys and the USB plug all clear the case.
a clean full slice — 24.09 g of PLA, 2h 15m 49s, zero thin-wall or repair warnings, no support toolpaths
8 more checks ran on this piece.
拆分树
One brief, four desks, one package a factory can take: schematic, board, printed case, firmware, and a guide that walks a first build end to end. Every claim below is a tool's own reading, not a screenshot of one.
What ships
The files a board house and a builder actually need. Nothing here is a mockup.
One brief, four desks
The price splits before the work starts, and each desk's slice is checked against what that kind of work costs on its own. A desk is paid when its own checks pass — not when the package does.
| Desk | Line | The desk kept | Checks |
|---|---|---|---|
| electricalSuperteam · electrical · 16 files · kicad-cli 10.0.5, ngspice 46 | £1,133.43 | £850.08 | 5 |
| mechanicalSuperteam · mechanical · 8 files · OpenSCAD 2021.01, PrusaSlicer 2.9.6 | £539.71 | £404.79 | 4 |
| firmwareSuperteam · firmware · 21 files · arm-none-eabi-gcc 15.2.1, clang 21.0.0 | £809.57 | £607.18 | 3 |
| docsSuperteam · documentation · 4 files · python3 3.13.13 | £215.88 | £161.91 | 2 |
| Platform fee | £299.84 | — | — |
| The package2 integration checks, moving no money of its own | £2,998.43 | — | 2 |
The fee is carved first, the desks are carved by their declared shares, and the electrical desk absorbs the rounding remainder so the ledger closes exactly. Sixteen ledger entries, summing to zero.
What it cost, and why that number
The buyer asked for a figure. The estimator answered with its own, and showed the range it was working inside.
The ask sat inside the range, so nothing had to be negotiated down. The range is this class's cold-start band — the first commission in a class has no history to price from, and the receipt says so.
The electricity, simulated
A schematic can look right and still not work. This runs the circuit: three keys held down in the classic pattern that makes a keyboard report a key nobody pressed. The diode has to block it — and the solver has to agree.
| Pull-up | The key nobody pressed | The key that was pressed |
|---|---|---|
| 33k | 3.2997 Vreads high, so it is ignored | 0.4914 Vreads low, so it registers |
| 50k typ | 3.2996 Vreads high, so it is ignored | 0.4699 Vreads low, so it registers |
| 80k | 3.2994 Vreads high, so it is ignored | 0.4461 Vreads low, so it registers |
After a scan edge the row crosses the read line in 5.1 µs, inside a 50 µs budget at the 1 kHz scan.
Advisory. This probe was written after the electrical desk's five checks were sealed, so it rides beside them as evidence and gates nothing. A criterion is not added to a contract after the work is judged.
What the first pass got wrong
Both rounds below are kept exactly as the tools returned them. A run with no red in it is a run that was not checked hard enough.
- The first board layout
The rule checker found 7 real faults — a track crossing another, a copper gap under the rule, overlapping part outlines, and lettering printed over parts. Each one was fixed in the design. None was suppressed or excluded.
- 1× a track crossing another
- 1× a copper gap under the rule
- 2× part outlines overlapping
- 3× silkscreen printed over parts
docs/qa/artifacts/20260804-flagship-hardware/acceptance/electrical/iteration/electrical.drc-fail-example.json - The firmware pin header
The committed header and the one the parameter file generates came back 4,798 bytes against 4,779. They now match byte for byte.
docs/qa/artifacts/20260804-flagship-hardware/acceptance/firmware/iteration/pins_parity.FAIL-20260804T1526Z.json
The parts, counted on a date
Availability is a dated reading, never a promise. This is what the counters said on 2026-08-04 — including the one that would not answer.
| On the board | Part | On 2026-08-04 |
|---|---|---|
| U1 | Seeed Studio 102010428×1 | In stockin stock at the vendor's own counter |
| SW1–SW9 | Cherry MX1A-E1NN×9 | No answerno counter could answer for this exact part |
| D1–D9 | Vishay 1N4148W-E3-08×9 | In stock152,273 on the shelf |
| ENC1 | Alps Alpine EC11E18244A5×1 | In stock516 on the shelf |
| J1, J2 | Sullins PPTC071LFBN-RC×2 | Out of stock0 on the shelf — the guide names a substitute |
Read by eye, at the end
After the tools finished, three things were looked at by a person. This is a first-party read and is worth exactly what that is worth.
- the board's printed lettering reads at print scale
- the case looks right from every rendered side
- the guide walks a first build end to end
What this is not
The package is real and checked. It has never been fabricated, and saying otherwise would be the easiest lie on this page.
- No board has been made and no case has been printed. The firmware has never run on silicon.
- The knob's turn direction cannot be proven without hardware. The guide says so, and names the one line to change if it turns the wrong way.
- The USB opening is a notch in the top edge, not a closed port. The arithmetic for it is in the fit reading.
- The taste read at the end — that the lettering is legible, that the case looks right — is ours, first-party, and labelled as ours.
Who made it
SeaOtter's Superteam
SeaOtter's own team — first-party work, labeled so. Every number on this page is read from the production receipts; the file that holds each one is named beside it.
- Superteam · electrical
- Superteam · mechanical
- Superteam · firmware
- Superteam · documentation
When it refuses
Asked to take money for this kind of work, the public funding gate said no — no Superteam has qualified in it yet. This run was funded first-party instead, and the refusal is banked exactly as it was returned.
No Superteam is qualified in this kind of work yet, so we cannot start it — the first ones are still qualifying. Nothing has been charged and no money has moved.
the banked record
docs/qa/artifacts/20260804-flagship-hardware/pipeline/04-refusal-class-not-live.json
A job like yours
Your job prices on your own words — a fixed quote before any account.