Week Aug 2 – Aug 8, 2026

The board meets the bench

One of the two machine designs was deleted, and the controller board was powered up for the first time with half of it working.

On 3 August the counter machine's tree was deleted and the tall narrow machine moved into its place. One content root, one edition. The two-edition machinery collapsed with it — the edition list, the commit hook, the path checker and the render tools all shed the machinery that needed two trees to exist. The published silhouette moved from 317 × 375 × 337.7 mm to 223 × 481 × 358 mm.

The same day, a delivered controller board was put on a bench and driven from a new console.

Working: the power rails, the microcontroller, USB-C flashing, WiFi at 15 networks, an RS485 loopback at 6 of 6, the buzzer, the status LEDs, the gas divider at 3020 mV, the compressor interlock, and both peristaltic pump drivers turning a real pump on both channels. A 4.3-inch touch panel on the board's RS485 link ran a pump remotely and held it to prime across 32 consecutive holds, with every one of 77 frames arriving.

Dead: the whole I²C bus — both I/O expanders, the real-time clock, all twelve valve outputs, ten reed inputs and the fan output.

Answered — 10 Unreachable — 6 power rails the microcontroller USB-C flashing WiFi — 15 networks RS485 loopback — 6 of 6 buzzer status LEDs gas divider — 3020 mV compressor interlock both pump drivers, on a real pump I/O expander 1 I/O expander 2 real-time clock 12 valve outputs 10 reed inputs condenser fan output All six sit on one I²C bus. The drill file carried 135 holes where the board had 152 vias — partial-span vias were dropped.
The main board's first hour powered up. The fab drilled exactly what it was sent.

The cause is the project's own drill-file generator, which emitted 135 holes against 152 vias because partial-span vias were dropped. The fab drilled exactly what it was sent.

Ten more assembled boards were ordered that day, $702.11, carrying four changes: all 152 vias drilled behind an assertion that matches every via one-to-one before any file is written, inner-layer annular rings on three connector pins, the error LED rotated, and the USB bridge swapped to a second-source part.

The appliance became one packed, machine-graded model. It went from 31 of 41 bodies placed against 11 gates to 77 of 77 bodies placed against 72 gates, four of which still fail. Two further models were born beside it — a standalone flavor-manifold placement study and a cold-core interior model. The enclosure build procedure was rewritten to match: eight cards deleted, nine written.

Nine commits that rebuilt the assembly around a new front-half module in one pass were reverted on 7 August, on the stated grounds that a pack assembled off bounding boxes is not something to build on. The module survived the revert, was rebuilt incrementally over the next two days, and became the assembly proper on the 8th.

The added laser-cut sheet-metal fire enclosure was designed out — eight files including a DXF, a STEP and its assembly card. The R-600a donor's own moulded cover remained over the terminal block and PTC module, and a thermal cutoff was added against that cover by a small printed clamp. The retained cover's qualification against the applicable enclosure provisions remains open. Per-unit parts cost fell from $1,346.86 to $1,296.40.

Two ledgers were born: attended human build time at 9 h 55 m per unit at $100/hr, and machine occupancy at 94.2 printer-hours per unit, the latter anchored on one measured print of 1142 g in 14 h 22 m.

Nothing was delivered this week and nothing was printed.

All updates