Hojicha44-Left

Hojicha44 — split columnar keyboard (case + plate phase)

Custom split columnar mechanical keyboard (44 keys + 2 encoders, nice!nano x2). Roadmap: KEYBOARD_PROJECT_ROADMAP.md. Current phase: PCB release v0.5.0 cut + fab-pointable; plate rev-8 gated (+ new z-stack gate); case/cover v4.1 generated + gated (thin top plate, key-field ledge, floor −6.35) with committed STEP solids; ZMK CI green. Board release evidence: generated/release/v0.5.0; order-time print geometry: generated/print/v0.5.0/MANIFEST.txt (STEP files verified in CI); renders: generated/renders/v0.5.0. Remaining before any batch: the first-article checklist (spec/first-article.md) — coupon-first, per its order policy. Layout geometry lives in layout/ergogen/config.yaml (frozen snapshot: layout/- FINAL/keyboard_layout_v7); each half has a placed+routed board file — see Schematic.

Repository layout (what every top-level thing is)

Layout

Schematic

Why two projects: a single flat sheet produces one netlist, so "Update PCB from schematic" would drop both halves into one board file. One project per half keeps each outline, netlist and DRC separate; the half-to-half link (TRRS_T/TRRS_R1 crossover) lives in the spec, not the netlist.

Each half differs from the other only in the TRRS TX/RX swap: D1→TRRS_T on the left, D1→TRRS_R1 on the right, for a straight cable. The crossover is copper, so both halves share identical UART pinctrl (TX=P0.06 RX=P0.08) — no firmware swap; see zmk/.

python3 pcb/generators/gen_schematic.py   # spec check + both sheets + audit + ERC

python3 pcb/generators/gen_pcb.py         # place both boards + containment + DRC gate

The schematic generator also asserts the two things ERC and DRC cannot catch:

• Socket rows are 15.24mm apart, pitch 2.54mm. Socket_Strips_2x12_P2.54mm_NiceNano numbers 1–12 down the TX/D1 edge and 13–24 down the RAW edge in the opposite direction, so pads 1 (D1) and 24 (RAW) share the USB end. The stock PinSocket_2x12 is row-major with 2.54mm row spacing and would land every MCU net on the wrong pad while still passing both checks — and the module would not fit.
• Reference designators. KiCad treats a refdes not ending in a digit as un-annotated, which makes every netlist/PCB export warn even though ERC is clean.

kicad-cli lives inside the KiCad bundle (/Applications/KiCad/KiCad.app/Contents/MacOS/kicad-cli, v10.0.6 here) — it is not on PATH by default. The generator finds it automatically.

Build notes (things that will bite you at assembly)

TRRS cable — buy a 90° right-angle one. The two jack mouths face each other across 22.40mm of mouth-to-mouth clearance (plastic faces; 20.42mm board-edge to board-edge, each mouth 0.99mm recessed behind its edge — silk tops at 21.40mm, still recessed). A molded straight plug protrudes ~12mm past the jack face, so the pair needs ~24mm and is a tight fit at best. A 90° plug protrudes only ~2–4mm axially because the boot exits perpendicular, so the current gap is ample. 90° is the community standard for splits anyway (less port torque). Size the cable port-to-port +2–5cm minimum — both cables exit upward and arch over the gap.

Always power-cycle in this order: plug/unplug the TRRS with both halves switched OFF and USB unplugged from both, then switch on. Switch OFF alone is insufficient — USB still powers the controller. Sleeve-GND mates at an angle during insertion, so connecting live can glitch TX/RX.

Don't treat TRRS as a daily link. splitkb's own nice!nano FAQ is blunt: "Do not use a cable between the halves… shorting connections when unplugging while the battery is still on… unbalanced charging when two batteries are connected." This design already isolates ring 2, which blocks the charging half; the shorting half is only mitigated by the power-off ordering above. TRRS here is a bring-up/debug fallback — move to 4-pin JST if you want it in normal use.

Battery: the 301230 goes directly under the module. This is correct. The cell's south 5mm sits over the antenna keepout circle, which looks alarming on the preview but is exactly what Lily58, Corne, Sofle, the whole splitkb Aurora series, and nicekeyboards' own docs do — none apply a cell keepout. The r=5 copper keepout is on our PCB and is honoured independently; the cell is a different layer. A PCB trace antenna has a null perpendicular to its plane, so a cell 3mm below the antenna end is in the null, not shadowing the main lobe. If BLE range ever disappoints, the documented upgrades are a 301228 (90mAh) or a 306080 between PCB and bottom plate with 9mm standoffs.

Regenerate

npx ergogen layout/ergogen -o layout/generated/v2 --svg --clean
node layout/generators/dump_points.cjs            # restores the JSON/YAML sidecars (4.2.1 no longer writes them)
python3 layout/generators/export_kle.py --check
node layout/generators/render_preview.mjs

PCB (place + route; generator copper + preserved GUI copper):

python3 pcb/generators/gen_schematic.py   # spec check + both sheets + guards + audit + ERC
python3 pcb/generators/gen_pcb.py         # place + route + containment + copper checks + DRC

gen_pcb.py owns the outline, footprints, nets, placement and generator copper (deterministic router gen_route.py, revision-tagged so stale copper is dropped, never silently kept). GUI copper is the other way round: (segment …), (via …) and (zone …) are preserved verbatim from the board file (net codes remapped by name, since codes are not stable across netlist exports). GUI routes survive regeneration (generator copper is revision-tagged, so a router rev drops only stale generator copper, never GUI copper). The build fails, rather than quietly producing a broken board, if preserved copper no longer lands on a pad/via/track — the usual result of moving a footprint without re-routing — or if it enters the r=5mm antenna keepout.

Previews

• Paste layout/- FINAL/keyboard_layout_v2.json into keyboard-layout-editor.
• Paste layout/ergogen/config.yaml into ergogen.xyz.
• layout/generated/v2/keymap.png (+.svg) is a labeled render with the keymap baked in — no Ergogen PNG export exists, so this script draws it from the KLE (ImageMagick required).
• Never cross-paste: the KLE importer rejects mid-row rotation by design.