Stawamus column bench · Order 1

Bench wiring and tests

One column on the TMC5240 dev kit. A NUCLEO-G474RE drives a TMC5240-BOB. The BOB drives one lead-screw column on a wooden jig (or the fully printed stand, step E2). A TBMA732 encoder sits on the motor's rear shaft. Use it for software bring-up, weight tests (including 1.4 vs 2.0 A run current), failure tests and encoder health. Parts and prices: parts list. 3D: jig model.

Read first. Change wiring only with both supply channels off and USB unplugged. Check + and − twice: a reversed rail destroys the capacitors and the BOB. Never connect or disconnect the motor while the BOB is powered. There is no E-stop: the supply's OUTPUT button is the stop, so keep it within reach.

What you have

Part	Qty	Note
Board
NUCLEO-G474RE	1	STLINK-V3E on board, USB Micro-B (CN1)
TMC5240-BOB	2	1 + 1 spare. 4.5–36 V, max 2.12 A RMS, SPI only, no bulk capacitor fitted
Watterott 201835 heatsink + Orion OD4010-24HB 24 V fan	2 + 2	On the BOB above 1.4 A
StarTech USB2CUB1M, USB-C to Micro-B	1	Nucleo to Mac
SparkFun 12794 M-F jumpers (20 × 150 mm), Sullins jumper shunts	1 + 2	Logic: 8, female end cut off and soldered into the BOB. Encoder: 6 (step L). The shunts stay unused: the BOB has no pins, L7–L8 get a wire bridge
Adafruit CAN Pal 5708	2	Later (CAN FD), not wired
Power and protection
Lab dual supply (lab's own)	1	CH2 = motor line, 24–30 V, limit 3 A. CH1 = fan, 24 V / 0.1 A. OUTPUT button = the stop
Fuse 3 A / 58 V MINI (0997003.WXN) + Keystone 3568 holder	3 + 1	In from the first power
Nichicon UHW1J222MHD, 2200 µF / 63 V	2	On the rail
Pololu #3777 shunt, 33 V	1	On the rail, on standoffs
Panasonic EEU-FR1H101, 100 µF / 50 V	2	At the BOB's R1/R2 + 1 spare
WAGO 221-415 (5-way) / 221-412 (2-way)	2 / 4	RAIL+ and RAIL− junctions / motor leads
Wire 1.5 mm² red + black, 0.5 mm² red + black, 4 mm banana plugs, 1.5 mm² ferrules, heat-shrink, cable ties	—	1.5 mm² = supply to rail; 0.5 mm² = soldered to the BOB
Column
17HS16-2004D motor	2	1 + 1 spare. Black + green = coil A, red + blue = coil B
Tr8x12 lead screw, 200 mm, + brass nut	2	1 + 1 spare. 12 mm per turn
Adafruit 1176 coupler 5→8 mm, Olimex KFL08 bearing	2 + 2	1 + 1 spare each
Pololu #2266 bracket + 2 mm shim (local)	2 + 1	Shim behind the leg
DeoxIT L260 grease	1	For the brass nut
Printed: shelf_stop, nut_block, tower, platform, peg	—	`docs/print/`. Print now
Local: two planks, wood screws, M3 × 25 / M3 × 30 / M4 × 20 + nuts, glue, clamp	—	Jig stand and tower
No wood: printed stand_base, stand_tube, stand_carrier, stand_post	1 each	Instead of the planks, shelf, nut block and tower (step E2). Same column parts; M3 × 6 (Pololu bag), M3 × 25 and M4 × 20 + nuts
Encoder
MPS TBMA732-Q-LT-00A	4	1 + 3 spares. MA732, 3.3 V only, A/B/Z 1,024 ppr out of reset. CN1 / CN2 come fitted: two red 8-way SMD receptacles, wires soldered on their legs (step L)
Radial 9049 magnet, 6 × 2.5 mm N35SH	5	Diametric, 150 °C max
Printed: encoder_magnet_holder, encoder_mount, encoder_tray	1 each	Holder: print now. Mount and tray: after measuring the rear shaft (`REAR_SHAFT_MM`) and the board thickness (`pcb_t`), step L
M2.5 × 5 nylon, × 8, × 6; 0.5 mm nylon washers; 28 AWG rainbow ribbon	—	4 screws each, 10 washers. Encoder leads: 4 ribbon wires + 2 cut M-F jumpers (step L)
Tools and weights
Biltema dial gauge 19-1443 + magnetic stand	1	On a steel plate ≥ 80 × 80 × 5 mm (local)
Weight discs 2.5 kg / 1.25 kg, kitchen scale 5 kg	1 / 2, 1	Biltema
From the lab	—	Meter, scope, soldering iron, crimper, thermocouple or IR thermometer, foam pad, pin or M4 bolt

Wiring

How each wire end is joined:S solder to the pin or padD Dupont: plug onto a pin or into a socket, no solderW WAGO: push in, close the lever (ferrule on 1.5 mm²)B banana plug: soldered on the wire, plugs into the supplyT clamped under the supply's terminalU USB cable· the part's own lead

CH2 / CH1 + − and GND 3.3 V SPI and enables encoder A/B/Z L7–L8 wire bridge ● joined · gap in a line = crossing, not joined

Everything on the bench in one picture. The 30 V path runs left to right along the top, logic from the Nucleo enters the BOB's left header, and motor and encoder leave its right header. BOB headers: L = left J201, R = right J202, pin 1 at the top. Exact wires and notes for each line: the table below. Where the boards sit around the jig, in 3D: jig model, tab "Wiring".

Build the 30 V rail

RAIL+ and RAIL− are two 5-way WAGO 221-415 lever connectors, not lengths of wire. Nothing is cut or chained: each part sits across the two WAGOs, so everything is in parallel. Only the 3 A fuse is in series, in the CH2 + lead.

Openings 1–5 of each WAGO: everything that enters a WAGO is W; soldered joints are S. C1, C2 = 2200 µF / 63 V, striped leg = minus. Opening 5 feeds the BOB; the 100 µF sits on the BOB's own pins.

Opening	RAIL+	RAIL−
1	Lead from the 3 A fuse (CH2 +)	CH2 − lead
2	C1 +	C1 − (striped leg)
3	C2 +	C2 − (striped leg)
4	Shunt + (VIN)	Shunt − (GND)
5	0.5 mm² red to BOB R1 +VM	0.5 mm² black to BOB R2 GND

Part	How
2200 µF caps	Legs straight into the WAGOs, not shorter than ≈ 10 mm. Stripe = minus. Reversed, a cap can burst.
33 V shunt	Only + and −. It sits across the rail and clamps above 33 V.
WAGOs	10–20 cm from the BOB.
100 µF	Soldered directly on the BOB's R1 (+VM) and R2 (GND), stripe to R2, shortest legs, together with the two rail wires.
Motor	Short 0.5 mm² wires soldered to R3–R6. Each motor lead joins its wire in a WAGO 221-412: black R4 A1, green R3 A2, red R6 B1, blue R5 B2. Wrong direction: `params --set invert_dir_mask=1` (column disabled).
Rules	Never plug or unplug the motor with power on. First power at 24 V with the CH2 limit at 0.5 A, then 30 V. Never above 30 V.

Every connection

The BOB has bare holes: every wire at the BOB is soldered S. Logic leads: an M-F jumper with its female end cut off, D at the Nucleo. Encoder VDD / GND: female end D on the Nucleo's morpho pin.

#	From	To	How	Wire	Note
Motor line (CH2)
P1	CH2 +	Fuse holder	B→S	1.5 mm² red, red banana plug	3 A / 58 V MINI in a Keystone 3568: solder, heat-shrink both pins, tie down. Not a 32 V car fuse.
P2	Fuse holder, other pin	RAIL+ (WAGO 221-415)	S→W	1.5 mm² red, ferrule
P3	CH2 −	RAIL− (WAGO 221-415)	B→W	1.5 mm² black, black banana plug, ferrule	Don't link CH1 − and CH2 − at the supply.
P4	C1 +, C2 + / C1 −, C2 −	RAIL+ / RAIL−	·→W	Cap legs straight in, ≥ 10 mm	2 × 2200 µF / 63 V. Stripe = minus, to RAIL−.
P5	Shunt VIN / GND	RAIL+ / RAIL−	S→W	1.5 mm², short	Pololu #3777 on standoffs, away from caps and wood.
P6	RAIL+	BOB R1 +VM	W→S	0.5 mm² red, soldered at the BOB	Never above 30 V (BOB max 36 V).
P7	RAIL−	BOB R2 GND	W→S	0.5 mm² black, soldered	Star point at the BOB. The only other ground is the Nucleo's, through L2.
P8	100 µF / 50 V	R1 (+) and R2 (−), at the pins	·→S	Soldered on the pins, shortest legs	Stripe to R2, together with P6 and P7. The BOB has no bulk cap (C209 not fitted).
Motor (BOB right header)
M1	R4 A1	Motor black	S→W	0.5 mm² soldered at the BOB, WAGO 221-412 to the motor lead	Coil A
M2	R3 A2	Motor green	S→W		Coil A
M3	R6 B1	Motor red	S→W		Coil B. Wrong direction: `params --set invert_dir_mask=1`, column disabled.
M4	R5 B2	Motor blue	S→W
Logic (Nucleo → BOB left header)
S1	CN6-4 3V3	L1 VCC_IO	D→S	M-F jumper, female end cut off: soldered into the BOB hole, male end into the Nucleo
S2	CN6-6 GND	L2 GND	D→S		Logic ground
S3	CN5-3 D10, PB6	L3 CSN	D→S		Idle high
S4	CN5-6 D13, PA5 (SPI1)	L4 SCK	D→S		Also drives LED LD2: don't use LD2
S5	CN5-4 D11, PA7 (SPI1)	L5 MOSI	D→S
S6	CN5-5 D12, PA6 (SPI1)	L6 MISO	D→S
S7	L8 CLK	L7 GND	S→S	Bridge wire (a cut capacitor leg), bottom side	CLK to GND = internal clock. Must not float.
S8	CN5-1 D8, PA9	L9 DRV_ENn	D→S	M-F jumper, female end cut off, soldered	BOB pull-up: driver off until the MCU drives it low
S9	CN9-8 D7, PA8	L10 SLEEPn	D→S	M-F jumper, female end cut off, soldered	Optional (BOB pull-up). Low resets every register.
—	A0 / PA0, DIAG0 PA10, DIAG1 PB5	open	—	—	PA0 open reads 'pressed', hence `--no-estop`. DIAG: the firmware reads the flags over SPI.
—	L11 REFL, L12 REFR, R7 GND, R8 AIN, R9 OV	open	—	—	Keep free on the Nucleo: PC4/PC5, PB3, PA13/PA14 (SWD), PA11/PA12 (FDCAN1 later).
Encoder (TBMA732 → BOB), fitted in step L
E1	CN1.1 VDD (leg)	Nucleo CN7-16 3V3	S→D	M-F jumper (red, black), male end cut off: soldered on the leg, female end on the Nucleo's morpho pin	Never 5 V: MA732 abs max 4.6 V. CN7-18, right beside CN7-16, is 5 V. Same net as L1: nothing is cut or soldered at the Nucleo or the BOB.
E2	CN1.2 GND (leg)	Nucleo CN7-8 GND	S→D		Same net as L2
E3	CN1.8 TEST (leg)	CN1.2 GND (leg)	S→S	2 cm of white ribbon wire, insulated	MA732 datasheet: connect TEST to ground; the board leaves it open (user guide Fig. 9). Outside legs 4 and 6
E4	CN2.5 A (leg)	BOB R10 ENCA	S→S	28 AWG ribbon wire (yellow, green, blue, brown), soldered on the leg and in the BOB hole. The four are twisted together, away from the motor leads	1,024 ppr = 4,096 counts/rev out of reset
E5	CN2.4 B (leg)	BOB R11 ENCB	S→S
E6	CN2.6 Z (leg)	BOB R12 ENCN	S→S		Index, once per rev
E7	CN1.8 leg (GND through E3)	BOB R7 GND	S→S		Return path along A / B / Z. On leg 8 so that no leg takes more than two wires
—	CN1.3–1.7 (–, CS, SCLK, MISO, MOSI); CN2.1–2.3 (SSCK, SSD, PWM); CN2.7 MGL, CN2.8 MGH	open	—	—	MGL / MGH: meter only, at gap setup (both low at the set gap). Beep check of every used leg: step L
Fan (CH1) and USB
F1	CH1 + (24.0 V, limit 0.1 A)	Fan red	T→·	Fan lead	Only for runs above 1.4 A. Never on the 30 V rail (fan max 26.4 V).
F2	CH1 −	Fan black	T→·	Fan lead
U1	Nucleo CN1 (Micro-B)	Mac	U→U	StarTech USB-C to Micro-B	Powers the Nucleo, flashing, serial at 921,600 baud

The stop and the protection

Part	Setting or behaviour
Stop	The supply's OUTPUT button, within reach of where you stand at the jig. It must switch CH2; if the supply has one button for all outputs, use that.
Over-current	CH2 limit 3.0 A is the first fuse. The 3 A / 58 V fuse protects the wiring if the limit is set wrong.
Over-voltage	2 × 2200 µF + 33 V shunt on the rail, 100 µF at the BOB. Supply OVP 35.0 V: above the shunt's 32.0–34.0 V, below the driver's 36 V.
Firmware	`--no-estop` on every command (PA0 not wired). STREAM mode: 200 ms without a frame sends the column to rest height. MANUAL mode (all bench tests): a running move finishes (≤ 120 mm, ≤ 0.6 m/s), then holds. There the button is the only stop.
After a stop	The caps hold R1 up ≈ 0.1 s (estimate). The 10 ms health poll then sees undervoltage, disables the column and drops its reference. To continue: CH2 on, then `init`, `enable`, `home`.
Power order	On: USB, then CH2. Off: `disable`, CH2 off, CH1 off, USB last.

Facts the firmware uses

Pin map and limits, from firmware/board/board_g474.h

MCU            STM32G474RE, 24 MHz HSE
SPI1 TMC5240   mode 3, 40-bit, 170 MHz / 32 = 5.3 MHz (/64 if errors)
               SCK PA5, MISO PA6, MOSI PA7, CS PB6
DRV_ENn        PA9, push-pull (BOB pull-up: off until low)
SLEEPn         PA8, held high (low wipes every register)
DIAG0 / DIAG1  PA10 / PB5, internal pull-ups, active low (not used yet)
E-stop         PA0, internal pull-up, high = stop; bench: not wired, --no-estop
Motion         chip ramp in velocity mode follows the 1 ms planner;
               positioning mode at rest; homing with SW_MODE.sg_stop
Microsteps     MRES 64 + intpol = 12,800 per rev (AMAX limit at 256)
Encoder        TBMA732 (MA732) A/B/Z, 1,024 ppr, no SPI or setup
               boots at enc_counts_rev 4096: binary, ENC_CONST 0x00032000
               (3.125); reversed: encoder_invert_mask=1 (0xFFFCE000)
Encoder lag    MA732 8-10 us = about 2 counts at 50 rev/s; ABZ
               hysteresis 0.52 deg = 6 counts (limit below: 41 counts)
Lost steps     ENC_DEVIATION (0x3D) = 128 (2 full steps), ENC_STATUS read
               every 1 ms -> event ENC_DEVIATION
               GCONF.stop_enable = 0 (else ENCA high stops the motor)
Limits         accel <= 7,857 mm/s^2, speed <= 600 mm/s
Current        RREF 12k: max 2.12 A RMS; cap 1.4 A after every boot; watch otpw
Link to Mac    LPUART1 PA2 / PA3 via ST-LINK VCP, 921,600 8N1, 0483:374E
Flash          ST-LINK, app at 0x08000000, no bootloader
Keep free      PA11 / PA12 (FDCAN1), PA13 / PA14 (SWD)

Steps, in order

Firmware: firmware/dist/firmware_nucleo_g474re_cube.bin (STM32Cube build; ping shows a build id ending in -cube). It has run on a bench: boot, the ST-LINK link, SPI-OK, init read-back and 120 mm strokes at 0.8 A. Every command carries --board nucleo-g474re --no-estop: it finds the ST-LINK port at 921,600 baud, and no E-stop is wired. bench = .venv/bin/bench from the repository root (or run source .venv/bin/activate once: then the blocks below paste as they are; it is python -m link, the code in software/link). Details: firmware/README.txt section 12. Ticks stay in this browser.

Prepare, everything unpowered

You will: solder S the BOB wires, the 100 µF, the fuse holder, the banana plugs and the shunt leads; crimp ferrules and push into WAGOs W; plug the jumpers' male ends into the Nucleo D. Every wire at the BOB is soldered. Soldering first, plugging last. Beep-check in B before any power.

BOB, solder first S. The BOB has bare holes, no pins, so every wire is soldered straight into its hole (more reliable than a plug):
1. Power and motor, R1–R6: 0.5 mm² red to R1, black to R2, the 100 µF on R1 (+) / R2 (stripe), four 0.5 mm² wires, ≈ 10 cm, to R3–R6. Heat-shrink.
2. CLK, L7–L8: no shunt (it needs pins). Bridge L7 to L8 with a short bare wire (a cut capacitor leg) on the bottom side.
3. Logic, L1–L6, L9, L10: cut an M-F jumper in half, strip 3 mm, solder the cut end into the hole. Its male end plugs into the Nucleo socket later.
4. Encoder, R10–R12 and R7: the encoder leads (step L) solder straight into these holes, no jumper.
Push each wire in from the top, solder underneath, trim. Tug-test every joint.
Motor current never goes through a Dupont. If a logic wire must come off again, cut it and re-solder.
Rail, rows P1–P8 (Build the 30 V rail). Solder first S: banana plugs B onto the 1.5 mm² supply leads; the fuse holder between the red lead and the RAIL+ lead, both pins heat-shrunk and tied down; 1.5 mm² leads onto the shunt's VIN and GND. Then ferrules on the 1.5 mm² ends and everything into the two WAGO 221-415 W: cap legs straight in, stripe to RAIL−. Shunt on standoffs, away from caps and wood. 3 A fuse in.
Motor, rows M1–M4, no soldering at the motor: each motor lead and its BOB wire into one WAGO 221-412 W. Step K opens them. The motor leads are thin: fold the stripped end double, then tug-test.
Logic, rows S1–S9: the leads are already soldered into L1–L6, L9 and L10 (A1). Push each male end into its Nucleo socket D. A0 (PA0) stays open.
Supply in INDEPENDENT mode, outputs off. Check CH2 at 30.0 V with a meter, OVP 35.0 V if it has one, then set CH2 to 24.0 V / 0.5 A for first power. CH1 24.0 V / 0.1 A, fan not yet connected. Series mode shorts a channel. CH2 never above 30 V: the shunt starts at 32.0 V and a voltage held above it destroys it.

Meter checks, unpowered

no solderingYou will: beep and measure with the meter. Everything still unpowered.

RAIL+ to RAIL− not shorted (the reading climbs as the caps charge). Fuse beeps end to end.
R3–R4 and R5–R6 ≈ 1.1 Ω each; R4 to R5 open; no coil to the motor body. If not, stop.
BOB L1 (3.3 V) not shorted to L2 (GND); R1 not shorted to L1.

First power at 24 V: flash, ping, registers

no solderingYou will: plug USB U, switch CH2 on, run commands.

USB first (VCC_IO up, DRV_ENn stays high), then CH2 on. 24 V at R1, no current beyond idle, nothing moves. Hand near the OUTPUT button.
Flash: cp -X firmware/dist/firmware_nucleo_g474re_cube.bin /Volumes/NOD_G474RE/. It reconnects when done; there must be no FAIL.TXT on the drive. Use cp -X (not Finder, not plain cp): macOS otherwise writes a hidden ._ file that can make the flash fail ("Flash algorithm write command FAILURE", no reply to ping); just copy again with -X. If C3–C4 fail, suspect the wiring or hardware first: this image has run on a bench (firmware/README.txt section 2).
ping ends in tmc5240; regs says SPI-OK (RAMPMODE, XACTUAL, VMAX …). SPI-FAIL: is CH2 on? Then the L7–L8 bridge and the L3–L6 joints.
vsupply: VS ≈ 24 V, chip near room temperature.

First spin, motor on the table

no solderingYou will: run commands, motor on the table.

Shaft free. CH2 limit 2 A. current 0.8, enable, spin --revs 1 --rps 1.
After a minute the motor is warm, not too hot to hold; the BOB stays cool to the touch. disable.

Commands (C and D)

bench --board nucleo-g474re --no-estop ping
bench --board nucleo-g474re --no-estop regs
bench --board nucleo-g474re --no-estop vsupply
bench --board nucleo-g474re --no-estop current 0.8
bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop spin --revs 1 --rps 1
bench --board nucleo-g474re --no-estop disable

Build the column jig

no solderingYou will: screw and glue the jig. To fit the motor, open its four WAGOs W and close them again after.

3D: jig model, tab "Column jig". Print files: docs/print/ (PETG or PLA, 4 walls, 40–60 % infill, no supports). Power off and the motor WAGOs open while you mount it.

No wood? Build the printed stand (E2) instead of E1–E7, then carry on with F.

Stand: board (≈ 160 × 315 mm, ≥ 12 mm) screwed and glued upright onto a base plank; base clamped to the table.
Bracket: Pololu 2266 on the board with the 2 mm shim behind its leg (motor axis 28.4 → 30.25 mm from the board), plate on top at ≈ 80 mm. Motor from below, shaft up, 4 × M3 × 6 (included).
Coupler: Adafruit 1176, underside ≈ 17 mm above the motor face, set screw on the flat.
Shelf + KFL08: printed shelf centred on the screw axis, top ≈ 5 mm above the coupler. KFL08 flat face down, 2 × M4 × 20, finger-tight.
Screw: through the KFL08, 7 mm into the coupler. Tighten the coupler, centre and tighten the KFL08, grub screws last. Thin film of grease on the screw.
Nut block: brass nut in the centre hole, flange up, 4 × M3 × 25 low-head. Thread on until its back rests on the board; candle wax on the board.
Ruler taped next to the block. Motor WAGOs closed again.

No weights before the tower and the 30 V checks (H). Travel ≤ 115 mm above home.

Printed stand (no wood)

no solderingYou will: print four parts and bolt the column into them, instead of E1–E7. The heights match the wooden jig, so the coupler, the bearing and step L's encoder parts sit exactly as there.

Print files: docs/print/stand_*.stl, made by make_parts.py, which checks every fit: the motor bolts, the coupler, the screw, the full stroke and the encoder under the motor. PLA or PETG, 0.4 mm nozzle, 0.2–0.28 mm layers, 3 walls, 20–30 % infill, no supports. Every part fits 180 × 180 × 180 mm (Prusa Mini).

Part	How to print	Mass	Hours, Mini / MK4
`stand_base`	Upright, feet on the bed. The motor plate and the top plate print as 43 mm bridges	≈ 47 g	≈ 4.3 / 2.1
`stand_tube`	Upright, foot plate on the bed (177 mm tall)	≈ 50 g	≈ 4.6 / 2.3
`stand_carrier`	Flat, nut traps down, 40 % infill	≈ 11 g	≈ 1.0 / 0.5
`stand_post`	Upside down, platform on the bed	≈ 50 g	≈ 4.6 / 2.3
Total	Estimates; the slicer's figure counts	≈ 160 g	≈ 14.5 / 7

Hardware: the motor, coupler, KFL08, screw and brass nut from E; 4 × M3 × 6 from the Pololu bag (motor); 2 × M4 × 20 + nuts (the KFL08 bolts also clamp the tube to the base); 4 × M3 × 25 low-head + nuts (brass nut to the carrier). No wood, no glue.

How it works: the motor hangs under the base's motor plate, shaft up. The tube stands on the base. The KFL08 sits on the tube's foot plate, under the tube's floor, between two side walls. The square carrier holds the brass nut and slides in the square tube, so it can't turn. At home it sits on the tube's floor, which is the stall stop. The post is a square sleeve with the 100 × 100 mm platform on top. It rests loose on the carrier and goes up and down with it. Boxes go on the platform. The free space under the motor (37 mm, open underneath, the front side open) takes the encoder stack of step L for any rear shaft up to ≈ 20 mm.

Print and dry-fit: the carrier slides through the whole tube by hand without catching (0.4 mm per side) and can't turn in it. Same for the post's sleeve. Tight: sand the carrier's sides, or reprint it with the slicer's XY compensation at −0.1 mm.
Motor: from below into the base, shaft up, the Ø22 boss in the plate's hole, leads out at the side (−x; the encoder board points to the front, +y). 4 × M3 × 6 from above, through the screwdriver holes in the top plate.
Coupler: Adafruit 1176 on the motor shaft, underside 17 mm above the motor face (11 mm above the motor plate), set screw on the flat. Reach in through the open sides.
Bearing bridge: the tube onto the base's two pins. Slide the KFL08 in through the open front or back, flat face down on the foot plate, between the two side walls. 2 × M4 × 20 up from under the base's top plate (through the top plate, the foot plate and the KFL08), nuts on top of the KFL08, finger-tight. The slots take a bolt spacing of 35.8–38.2 mm.
Nut carrier: brass nut in the centre hole, flange up. 4 × M3 × 25 low-head down through the flange, M3 nuts in the traps underneath; tightening pulls each nut up into its trap, and the bolt tips stay inside. Thread the carrier onto the top end of the lead screw, about 30 mm down.
Screw: lower it, with the carrier on it, into the tube, the carrier square with the tube. Through the floor and the KFL08, then 7 mm into the coupler. Tighten the coupler, then centre and tighten the KFL08's M4 bolts. Grub screws last: turn the screw until they face an open side. Thin film of grease on the screw.
Platform: the post down over the screw onto the carrier. It only rests there. Turn the coupler by hand: the carrier runs up and down without catching and sits flat on the floor at the bottom.
Feet: clamp two pads to the table, or screw them down, or weigh them down. With 5 kg on top at full height, ≈ 7 N sideways tips it over (estimate). Ruler taped next to the tube. Motor WAGOs closed again.

Travel ≤ 100 mm above home on this stand (G's 115 becomes 100). F's first moves are unchanged. Home is the carrier on the tube's floor. Boxes up to 5 kg, centred on the platform, only after H. On this stand they replace the tower, the peg and the discs of step I (I1, I4).

First moves on the jig (24 V, no weights)

no solderingYou will: run commands.

Block halfway up by hand. enable, setpos 60, then moves to 80 and 40 mm at 10 mm/s. One screw turn = 12 mm.

Homing

no solderingYou will: run commands.

home at 30 mm/s (never below 18): the chip's StallGuard stop, no switch. Hand on the OUTPUT button. Runs to the end without stopping ('no_stall') or stops early: sgtune, then params --set sgt_home=… (lower = more sensitive).
Home 10 times, note the zero spread. Then full moves to 115 mm, raising speed slowly. End on the post: move 0, disable.

Commands (F and G)

bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop setpos 60
bench --board nucleo-g474re --no-estop move 80 --speed 10
bench --board nucleo-g474re --no-estop move 40 --speed 10
bench --board nucleo-g474re --no-estop params --set tcool_speed_mm_s=15
bench --board nucleo-g474re --no-estop home
bench --board nucleo-g474re --no-estop move 115 --speed 50    # raise slowly
bench --board nucleo-g474re --no-estop move 0 --speed 10
bench --board nucleo-g474re --no-estop disable

30 V and the stop, no weights

no solderingYou will: change the supply, measure, press OUTPUT.

CH2 off, set 30.0 V / 3.0 A (check with a meter), on. init, vsupply ≈ 30 V; shunt and caps cold.
Find the OUTPUT button that switches CH2 (or all outputs). Place the supply so you reach it from the jig without looking.
Scope on R1 (ground clip on R2 only), trigger at 31 V. Unloaded moves at 0.1 and 0.4 m/s: peak below ≈ 32 V.
Stop mid-move at 0.4 m/s, no load: press OUTPUT. The column stops on undervoltage, peak ≤ 34 V. Then CH2 on, init, enable, home.

Commands (H)

bench --board nucleo-g474re --no-estop init          # CH2 was off: driver re-initialised
bench --board nucleo-g474re --no-estop vsupply
bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop home
bench --board nucleo-g474re --no-estop move 115 --speed 100
bench --board nucleo-g474re --no-estop move 5 --speed 400     # press OUTPUT mid-move
# after the stop: CH2 on, then
bench --board nucleo-g474re --no-estop init
bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop home

Weight tower, dial gauge, weights

no solderingYou will: assemble the tower, set up the dial gauge, weigh.

3D: jig model, tab "Weights". Load path: weights → tower → nut → screw → KFL08.

Tower over the screw onto the nut block, 4 × M3 × 30. Platform on its spigot, peg on top.
Dial gauge on the steel plate, plunger up, touching the platform only in the last few mm. Approach the reference from above, then zero.
Weigh everything that moves. Force = (total moving mass − 0.064 kg) × 9.81.
Discs over the peg, pin through the cross hole above them, every time. Foam pad on the shelf post.
Stop test as H at 10 N, hands clear. OVP must not trip in normal moves (the shunt clamps first). Above 35 V: stop.

Load	Total moving mass	Roughly
2.5 N	0.32 kg	bolts or a water bottle
10 N	1.08 kg	tower + light load
25 N	2.61 kg	2 × 1.25 kg discs + tower
40 N	4.14 kg	2.5 + 1.25 kg discs + tower

Weight tests

no solderingYou will: stick on the heatsink; clamp the fan leads under CH1's terminals T; run tests.

Predictions from reports/BENCH_V1/PREDICTIONS.md (archived study: studies/ARCHIVE.md, tag pre-cleanup), made for the TMC2240; the TMC5240 has the same power stage. Cruise ≤ 0.6 m/s, ramp 4 m/s².

Test	How	Prediction	Wrong if
Light loads	2.5 N and 10 N: 20 × (lift 0→120, drop, step 60→70) at 0.4 m/s	No lost steps	Any stall flag, or gauge > 0.5 mm off
Timing	10 N: full lift and 10 mm step; firmware times + 240 fps video of the ruler	Lift ≈ 0.4 s, step ≈ 0.1 s	Lift > 0.5 s or step > 150 ms
Speed vs load	25 N and 40 N, cruise 0.1 → 0.6 m/s, 20 clean cycles per speed. Once at 1.4 A, once at 2.0 A (below).	2.0 A: 25 N clean to ≈ 0.4 m/s, 40 N to ≈ 0.3 m/s. About half the 40 N lost steps of 1.4 A.	2.0 A no better than 1.4 A at 40 N
Motor heat	10 min of cycles at 40 N at each current; motor case temperature at the end	Copper loss ≈ 4.3 W at 1.4 A, ≈ 8.8 W at 2.0 A	Case > 80 °C at 2.0 A
Stall detection	20 provoked stalls at 0.4 m/s, then 20 clean cycles at 10 N	≤ 2 of 20 missed, no false flags	More missed, or any false flag (re-tune first)
Return accuracy	Dial gauge after each block of 20 cycles	Within 0.5 mm	> 0.5 mm (lost steps or slipping coupler)
Friction (optional)	Slide test, cardboard box on the woven PTFE fabric, ≈ 5 kg and ≈ 10 kg	As measured at 0–575 g: static 0.23, kinetic 0.16	Friction clearly rises with load

Commands (J, at 1.4 A)

bench --board nucleo-g474re --no-estop a1 light     --mass-kg 0.32 --cycles 20   # 2.5 N
bench --board nucleo-g474re --no-estop a1 light     --mass-kg 1.08 --cycles 20   # 10 N
bench --board nucleo-g474re --no-estop a1 timing    --mass-kg 1.08
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 2.61               # 25 N, 0.1 -> 0.6 m/s
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 4.14               # 40 N
bench --board nucleo-g474re --no-estop a1 stall     --mass-kg 1.08 --speed 400
bench --board nucleo-g474re --no-estop a1 return    --mass-kg 1.08 --blocks 3

Log moving mass, speed, stall flags, gauge and the scope's rail peak (the tool asks). Keep falls short: a fast-falling load drives the motor as a generator into the shunt. A move runs to its target: OUTPUT is the only stop.

Run current: 1.4 A vs 2.0 A

Decides the production preset: 2.0 A only if it clearly loses fewer steps at 40 N and the case stays ≤ 80 °C. The firmware caps the current at 1.4 A after every boot.

Cooling: Watterott heatsink on the BOB's bottom side, right behind the TMC5240, between the header rows (no vias under the chip: stick it on the package instead). The BOB has no mounting holes: lay it fins up on a dry wooden or foam block with a groove, so the wires soldered into it leave sideways underneath. Fan on CH1 (rows F1–F2), 2–5 cm from the heatsink. Thermocouple on the motor case (or IR thermometer); note room temperature.
1.4 A (the default cap): speedload at 25 N and 40 N, then the 10 min heat run. Read the case temperature and vsupply's chip temperature at the end.
Raise the cap on purpose: params --set current_cap_ma=2000. RAM only: every reboot is back to 1.4 A. The firmware refuses anything above 2000 mA.
2.0 A: same runs. Let the motor cool to room temperature first.
Stop if the case passes 80 °C, status shows OTPW, or the fan stops. Afterwards set 1.4 A and the 1400 cap back (or reboot).

Commands (run current)

# 1.4 A
bench --board nucleo-g474re --no-estop current 1.4 --hold 0.8
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 2.61 --notes "1.4 A"
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 4.14 --notes "1.4 A"
bench --board nucleo-g474re --no-estop a1 light --mass-kg 4.14 --speed V --cycles N --yes --notes "1.4 A heat"
bench --board nucleo-g474re --no-estop vsupply
#   V = top clean 40 N speed at 1.4 A (100 if none); N = cycles for 10 min (time 20 first)

# 2.0 A (heatsink and fan on)
bench --board nucleo-g474re --no-estop params --set current_cap_ma=2000
bench --board nucleo-g474re --no-estop current 2.0 --hold 0.8
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 2.61 --notes "2.0 A"
bench --board nucleo-g474re --no-estop a1 speedload --mass-kg 4.14 --notes "2.0 A"
bench --board nucleo-g474re --no-estop a1 light --mass-kg 4.14 --speed V --cycles N --yes --notes "2.0 A heat"
bench --board nucleo-g474re --no-estop status

# back to the default (current first: the cap can't go below it)
bench --board nucleo-g474re --no-estop current 1.4 --hold 0.8
bench --board nucleo-g474re --no-estop params --set current_cap_ma=1400

Failure tests: does a loaded column hold when the motor stops?

no solderingYou will: open and close the four motor WAGOs W; run tests.

A. Mechanical hold: block halfway up, disable, CH2 off, rail below 2 V, then open the 4 motor WAGOs (no braking through the driver). Add 0.5, 1, 2, 3, 4 kg; note where it creeps down. Prediction: holds to ≈ 27 N (2.7 kg).
B. Driver off (WAGOs closed, after H): loaded at halfway, disable. Film against the ruler, scope on the 30 V line. 10 N first.
C. Power cut (after B): same, but press OUTPUT (CH2 off). The motor brakes the fall into the 30 V line, peak ≤ 34 V. The spare BOB is on hand if this one dies.

Commands (K)

bench --board nucleo-g474re --no-estop move 60 --speed 50
bench --board nucleo-g474re --no-estop disable          # A and B
# C: CH2 off instead; afterwards CH2 on, then
bench --board nucleo-g474re --no-estop init
bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop home

Encoder: lost steps and column health

You will: solder S six wires and the TEST bridge onto the legs of the TBMA732's two red connectors, before the board goes into its tray, then beep every used leg against the MA732 before any power. VDD and GND: M-F jumpers, female end D onto the Nucleo's CN7-16 and CN7-8. A, B, Z and the spare GND: 28 AWG ribbon wires, soldered S into the BOB holes R10, R11, R12 and R7.

The TBMA732 on the rear shaft feeds the BOB's encoder inputs (rows E1–E7). The chip compares it with its own position every 1 ms and raises ENC_DEVIATION at 2 full steps. 3D, step by step: jig model, tab "Encoder". The model still shows bare pads and the old leads; the legs sit on those same pads.

CN1 and CN2 come fitted. Each is a red 8-way SMD receptacle: two staggered rows of contacts at 1.27 mm, 4 gull-wing legs on each side. The user guide lists Würth 690367280876 or TE 7-2178711-8 for this footprint. Red with a pick-up cap points to the Würth part, since the TE part is black. Each leg is soldered on its own pad, so leg n = pin n of the guide's Fig. 12: odd legs on one side, even legs on the other, and pin 1 is the odd leg at the stem end. The loose red pieces were the pick-up caps.

Wires on the legs

Ribbon (Assmann AWG28-10/F/300): peel off yellow, green, blue and brown, long enough to reach the BOB with slack (30–60 cm). Also cut 2 cm of white for the bridge. Strip 1.5 mm and tin.
Jumpers: 2 M-F, a red and a black if the pack has them. Cut them near the male end so ≈ 14 cm of wire stays on the female end.
Solder each wire on top of its leg's foot, where it lies on the pad (figure). Use a fine tip and flux, about 1 s per joint, because the housing is plastic. Each leg takes one wire, except legs 2 and 8 of CN1, which also take the bridge.
Inner legs (CN1.1, CN2.4, CN2.6): run the wire up and over the connector, not across the Ø3 hole, because its screw goes in later.
Bridge: the white piece from CN1 leg 2 to leg 8, outside the feet of 4 (CS) and 6 (MISO). Twist yellow, green, blue and brown together.

Mount (printed)

Print now: encoder_magnet_holder (it doesn't depend on the shaft length).
Print after measuring: encoder_mount and encoder_tray (34 × 84.6 × 6 mm, flat, pocket up). Measure the rear shaft (REAR_SHAFT_MM) and the board thickness (pcb_t, 1.6 mm assumed). Set both in print/make_parts.py and run it; it checks the clearances.
Magnet into the holder with a drop of superglue, holder on the rear shaft.
Mount: 2 × M2.5 × 5 nylon into the 19.05 mm pair. Tray: 2 × M2.5 × 8 from below. Board chip up in the pocket: 2 × M2.5 × 6.
Air gap 1.5 mm, magnet to chip (≈ 85 mT, estimate). Each 0.5 mm washer between tray and walls adds 0.5 mm. The connectors stand ≈ 5.3 mm above the board, 38 mm outside the motor, so nothing on the jig comes near them (make_parts.py checks this). Keep 34 mm clear in front of the motor for the connector end and its wires.

S soldered: on the legs, the TEST bridge, and in the BOB holes R7, R10–R12 · D female end onto the Nucleo's CN7-16 / CN7-8, no solder

TBMA732 board end from the chip side, stem up, at 20 px per mm. Board and pin positions come from the TBMA user guide Rev 1.2 (Fig. 11, 12, Table 5); the connector comes from the Würth 690367280876 drawing (body 5.2 × 12.16 mm, legs 7.0 mm across). The leg colours show which legs to use. The wire colours are a suggestion; any distinct colours will do if you note them. CN2.1–2.3 are the MA732's SSCK, SSD and PWM, unused here. The inset shows the MA732 pins for the beep check, from the MA732 datasheet's top view (pin 1 at the dot, then anticlockwise).

Beep check, before any power

Use the meter on continuity, with nothing plugged in. Touch each used leg, then the MA732 pin it must reach. The QFN pins are 0.5 mm apart at the chip's edge, so touch the solder fillet at the side with a sharp probe, or a needle held against it. For VDD and GND, the two ends of C1 are easier. The check is meter-only and can't harm the board. If any leg beeps to a different pin than listed, stop and don't power it: the legs are numbered differently from the user guide.

Leg	Wire	Must beep to	Must stay silent to
CN1.1 VDD	red jumper (E1)	MA732 pin 13, one end of C1	CN1.2: a short chirp is C1 charging, a steady beep is a short
CN1.2 GND	black jumper (E2), bridge	MA732 pin 8, the other end of C1	CN1.1
CN1.8 TEST	bridge (E3), brown (E7)	MA732 pin 10; CN1.2 once bridged	CN1.1
CN1.4 CS, CN1.6 MISO	none	—	CN1.2 and CN1.8 (the bridge stays clear of them)
CN2.5 A	yellow (E4)	MA732 pin 2	CN1.1, CN1.2
CN2.4 B	green (E5)	MA732 pin 6	CN1.1, CN1.2
CN2.6 Z	blue (E6)	MA732 pin 3	CN1.1, CN1.2
CN2.7 MGL, CN2.8 MGH	none (meter points)	MA732 pins 11 and 16 (optional)	—
Each wire's far end		its own leg	its neighbours in the twist

Option to buy: a plug instead of soldering on the legs

Würth's datasheet for the WR-MM male IDC 690157000872 (8-way, 28 AWG ribbon at 1.27 mm) lists 690367280876 as a mate. Digi-Key 732-1990-5-ND costs $0.90, with a 15-week lead time when checked on 2026-10-17. Digi-Key offers TE 215083-8 (Micro-MaTch male IDC, 28 AWG) as its substitute, with 813 shown in stock the same day. It uses the same Micro-MaTch interface, but no maker's table pairs it with this socket, so dry-fit it unpowered first. One plug per connector, pressed in a vise onto 8 conductors of the same ribbon, conductor 1 on pin 1. TEST to GND then joins conductors 8 and 2 within a few cm of the CN1 plug. Run the same beep check through the plug.

Power off. Board wired and beep-checked (table above) and mounted (cards above), air gap 1.5 mm. Then, with USB in and CH2 off, MGL and MGH both read low on a meter.
On 3.3 V alone (E1 and E2 plugged in, E4–E7 not yet soldered to the BOB), A toggles 0 / 3.3 V when you turn the shaft. Then solder rows E4–E7 into the BOB. The TEST bridge is already on.
USB, then CH2. With the driver off, turn the screw one motor turn by hand (12 mm): in regs, X_ENC changes by ≈ 12,800 (4,096 counts × ENC_CONST 3.125 = one turn in microsteps).
params --set encoder_mask=1 (column disabled), enable, home. If the error grows with travel: disable, encoder_invert_mask=1, enable, home.
A test move, then the health schedule into SQLite; analysis with deck-health report (python -m link.health).

Commands (L)

bench --board nucleo-g474re --no-estop regs
bench --board nucleo-g474re --no-estop disable
bench --board nucleo-g474re --no-estop params --set encoder_mask=1
bench --board nucleo-g474re --no-estop enable
bench --board nucleo-g474re --no-estop home
#   encoder error grows with travel: disable, params --set encoder_invert_mask=1, enable, home
bench --board nucleo-g474re --no-estop testmove stroke --speed 50 --reps 2
bench --board nucleo-g474re --no-estop health-run --db jig.sqlite --encoder
deck-health report --db jig.sqlite

CAN FD bench test

Before Order 2, with what is on the bench: one NUCLEO-G474RE, both CAN Pals from Order 1, no PC adapter. FDCAN1 talks to FDCAN2 over a short real bus, in both directions, with the normal bench image. The BOB stays wired and nothing moves. Command and output: firmware/README.txt.

Part	Qty	Status
Adafruit CAN Pal 5708 (TJA1051T/3, 2 × 60 Ω terminator with a switch)	2	have (Order 1)
120 Ω terminators	—	not needed: on the CAN Pals (switch ON)
Twisted pair + a ground wire, ≈ 30 cm	1	have: twist 24 AWG solid wire (Order 2) by hand
Jumpers: 2 M-F, 6 F-F	8	have (Order 1 M-F, Order 2 F-F)

How each end is joined:D Dupont jumper end on a pin or in a socketT CAN Pal screw terminalU USB cableS only the CAN Pals' headers (step 1)

CAN_H CAN_L GND 3.3 V TX / RX

Each CAN Pal runs on 3.3 V (it makes its own 5 V for the bus), so its TX / RX are 3.3 V like the Nucleo's. Power comes from the two IOREF pins: on this Nucleo IOREF is 3V3 (solder bridge SB39, on by default), because CN6-4 and CN7-16 already feed the BOB and the encoder.

#	From	To	How	Wire	Note
CAN Pal A on FDCAN1
C1	Nucleo CN6-2 IOREF (3.3 V)	A Vcc	D→D	M-F jumper, male in the Nucleo socket	IOREF = 3V3 (SB39 on)
C2	Nucleo CN6-7 GND	A GND	D→D	M-F jumper, male in the Nucleo socket
C3	Nucleo CN10-12 PA12 (FDCAN1 TX)	A TX	D→D	F-F jumper	MCU TX → transceiver TXD
C4	Nucleo CN10-14 PA11 (FDCAN1 RX)	A RX	D→D	F-F jumper	Transceiver RXD → MCU RX
CAN Pal B on FDCAN2
C5	Nucleo CN7-12 IOREF (3.3 V)	B Vcc	D→D	F-F jumper
C6	Nucleo CN10-20 GND	B GND	D→D
C7	Nucleo CN10-30 PB13 (FDCAN2 TX)	B TX	D→D		Free on the bench
C8	Nucleo CN10-16 PB12 (FDCAN2 RX)	B RX	D→D		Free on the bench
The bus
C9	A terminal H	B terminal H	T→T	Twisted pair
C10	A terminal L	B terminal L	T→T	Twisted pair
C11	A terminal GND (middle)	B terminal GND	T→T	Wire along the pair	Common ground
—	Both SLNT	open	—	—	High = silent. Both TERM switches ON

CAN

Steps

You will: solder the two CAN Pal headers S, plug jumpers D, clamp the bus T, run two commands. No image swap, nothing unplugged at the BOB.

CAN Pals: solder the included header on both S (the terminal block comes fitted). Both TERM switches ON, SLNT open.
Wire C1–C11, USB out and CH2 off.
60 Ω check, nothing powered: ohmmeter from CAN_H to CAN_L at either terminal block reads ≈ 60 Ω. ≈ 120 Ω: one TERM switch is OFF. Open: a broken wire.
Build: cp -X firmware/dist/firmware_nucleo_g474re_cube.bin /Volumes/NOD_G474RE/ (the build with canbus-test; no FAIL.TXT afterwards). bench --board nucleo-g474re ping must show that new build.
Test: bench --board nucleo-g474re canbus-test prints CAN bus test 500k/4M BRS: FDCAN1->FDCAN2 1000/1000 ok, FDCAN2->FDCAN1 1000/1000 ok, tx errors 0, rx errors 0, bus-off 0 On a failure a second line starts likely cause:; the table below says what to check. -v adds the counts per direction.
If the FD test fails: bench --board nucleo-g474re canbus-test --classic runs classic CAN at 500 k. Classic ok but FD not: the data phase (timing or cable). Both fail: the wiring.

The test says	Check
`likely cause: no ACK`	Termination (both switches ON, 60 Ω), 3.3 V on each Vcc, SLNT open
`likely cause: bus-off`	CAN_H / CAN_L swapped at one CAN Pal, or TX / RX swapped (PA12 / PB13 → TX, PA11 / PB12 ← RX)
`errors in the data phase only`	Timing or the cable at 4 Mbit/s: a clean `--classic` pass confirms the wiring
`errors in the arbitration phase`	Wiring: CAN_H / CAN_L, a loose contact, ground, termination
Anything else	3.3 V at each Vcc; GND C2, C6, C11; 60 Ω; SLNT not pulled high.

Commands (CAN)

bench --board nucleo-g474re ping                     # the new build
bench --board nucleo-g474re canbus-test              # 1000/1000 ok both ways
bench --board nucleo-g474re canbus-test --classic    # 500k classic only, if FD fails
#   --frames N: a longer or shorter run
# no --no-estop and no disable here: the motor supply can stay off, and --no-estop would
# refuse ("driver not initialised") because it re-arms the driver, which needs motor power

PC-side CAN (talking to the Mac) needs a USB CAN FD adapter: PEAK PCAN-USB FD, or a cheaper candleLight / slcan adapter (being evaluated). Not needed for the single-board pilot, which uses the debug serial port.

HIL test: a whole Tile-20's work on the bench

After the CAN FD test passed, with the same wiring (the BOB, the encoder, both CAN Pals C1–C11). A separate image runs 20 columns: the real motor is column 0, the other 19 are simulated inside the Nucleo, but their SPI traffic really goes out on the buses. The Mac's heights travel over the real CAN bus (FDCAN2 → FDCAN1) among a whole row's traffic, as on the deck. A recorded simulator run drives it. Nothing to wire. Details: firmware/README.txt section 17.

Hand on the supply. The motor follows a busy deck column at up to 400 mm/s over the full 0–120 mm. Free shaft only (nothing on the nut block). Not on the jig until a free-shaft run has passed; then --scale 0.9. Motor supply 29 V, run current 0.8 A first. Keep a hand on the supply's OUTPUT button for every run.

HIL

Steps

You will: flash the HIL image, run 2 minutes, then 30 minutes, then flash the bench image back. Each run ends with HIL PASS or HIL FAIL.

Dry run on the Mac: .venv/bin/python sim/tools/hil_replay.py run --virtual --duration 120 ends with HIL PASS. No hardware. It also proves the recording is there (else: hil_replay.py record, about 75 s).
Column clear: free shaft (not the jig yet). CH2 at 29.0 V, output OFF.
Flash: cp -X firmware/dist/firmware_nucleo_g474re_hil_cube.bin /Volumes/NOD_G474RE/, no FAIL.TXT.
Supply ON, then bench --board nucleo-g474re-hil --no-estop ping shows -cube-hil and columns 20/20; ... regs --col 0 says SPI-OK.
2 minutes: the replay below with --duration 120. Watch the shaft and the supply's current. It sets the limits (400 mm/s, 4 m/s²) and 0.8 A, checks 2 s idle (it refuses on any SPI timeout and prints which bus), enables all 20 columns, streams, then prints the report. If the real column strays more than 0.12 mm from its command, the firmware switches its driver off within 20 ms and the run stops.
30 minutes (only after a 2-minute PASS): --duration 1800. Feel the motor case at 10 and 30 min.
Back to the bench image: cp -X firmware/dist/firmware_nucleo_g474re_cube.bin /Volumes/NOD_G474RE/

PASS needs	Limit
Planner tick (without the simulated chips' own work)	no deadline miss, p99 ≤ 600 µs, max < 1 ms
CAN targets	0 lost, 0 duplicates, no bus-off
Real column vs. the firmware's command	≤ 0.12 mm (2 full steps), encoder the same
Watchdogs, SPI	no trip, no timeout

Commands (HIL)

.venv/bin/python sim/tools/hil_replay.py run --virtual --duration 120        # dry run, no hardware
cp -X firmware/dist/firmware_nucleo_g474re_hil_cube.bin /Volumes/NOD_G474RE/
bench --board nucleo-g474re-hil --no-estop ping                              # -cube-hil, columns 20/20
bench --board nucleo-g474re-hil --no-estop regs --col 0                      # SPI-OK
.venv/bin/python sim/tools/hil_replay.py run --no-estop --current 0.8 --duration 120 --json runs/hil/bench_120s.json
.venv/bin/python sim/tools/hil_replay.py run --no-estop --current 0.8 --duration 1800 --json runs/hil/bench_1800s.json
bench --board nucleo-g474re-hil --no-estop hil                               # the report again, any time
cp -X firmware/dist/firmware_nucleo_g474re_cube.bin /Volumes/NOD_G474RE/    # the bench image back
#   encoder fitted and wired on the BOB: add --real-encoder;  the jig (only after a free-shaft PASS): --scale 0.9

If it fails, the report says where: the tick figures, the CAN counters (check as in the CAN table above), or the real column's error. Keep the --json file and bench ... events.

Firmware: firmware/dist/firmware_nucleo_g474re_3x3_cube.bin (STM32Cube build). Tested in simulation only (its 1-column twin has run on a bench). Use --board nucleo-g474re-3x3, without --no-estop. Columns 0–8 = A1 A2 A3 B1 B2 B3 C1 C2 C3. It boots with every DRV_ENn high. Every pin below is checked against firmware/board/board_g474.h and UM2505 Table 16.

Wiring

30 V / 24 V + − and GND 3.3 V logic encoder A/B/Z 16-way ribbon ● joined · gap in a line = crossing, not joined

The whole rig. 30 V runs from the supply through F0 and K1 to the star point, then one fused branch per row. Each row board takes a ribbon from the carrier and feeds its three BOBs. The dashed box shows one column; all nine are wired the same way. IDs match the tables below. BOB headers: L = left J201, R = right J202, pin 1 at the top.

Build the star point

STAR+ and STAR− are two linked WAGO 221-415 each. The caps and the shunt sit across them in parallel. Only F0, K1 and the row fuses are in series.

Openings 1–5 of each WAGO. 2200 µF / 63 V: the striped leg is minus. Short 1.5 mm² links join a5 to b1.

Opening	STAR+ a	STAR+ b	STAR− a	STAR− b
1	From K1 terminal 8 (XP)	Link from a5	Supply − (P2)	Link from a5
2	Cap 1 +	Row A fuse (P6)	Cap 1 − (stripe)	Row A − WAGO (P8)
3	Cap 2 +	Row B fuse (P6)	Cap 2 − (stripe)	Row B − WAGO (P8)
4	Shunt VIN	Row C fuse (P6)	Shunt GND	Row C − WAGO (P8)
5	Link to b1	spare	Link to b1	spare

Each row has two more WAGO 221-415: row + takes the fuse lead (1) and R1 of BOB 1–3 (2–4); row − takes the STAR− lead (1) and R2 of BOB 1–3 (2–4). Opening 5 is spare. The 24 V side has two more: 24 V+ and 24 V−.

Every connection

How each wire end is joined:S solder to the pin or padD Dupont: plug onto a pin or into a socket, no solderW WAGO: push in, close the lever (ferrule on 1.5 mm²)B banana plug: soldered on the wire, plugs into the supplyT clamped in a screw terminalU USB cableR 6.3 mm receptacle, crimped, pushed on the relay tab· the part's own lead

#	From	To	How	Wire	Note
30 V power
P1	Supply +	F0 holder	B→S	1.5 mm² red, banana plug	10 A / 58 V MINI in a Keystone 3568, soldered and heat-shrunk, tied down
XP	F0 out	K1 2; K1 4 → K1 6 (link); K1 8 → STAR+ a1	S→R	1.5 mm², 6.3 mm blue receptacles	Both relay contacts in series
P2	Supply −	STAR− a1	B→W	1.5 mm² black, banana plug, ferrule
P3	STAR± a5	STAR± b1	W→W	1.5 mm², short	Links the two WAGOs of each star
P4	Cap 1, cap 2 (2200 µF)	STAR+ / STAR− a2, a3	·→W	Legs straight in, ≥ 10 mm	Stripe to STAR−
P5	Shunt VIN / GND	STAR+ / STAR− a4	S→W	1.5 mm², short	Pololu #3777 on standoffs, away from the caps
P6	STAR+ b2, b3, b4	Row A, B, C fuse IN	W→S	1.5 mm² red, soldered to the holder pins	5 A MINI, on the row board's 30 V zone
P7	Row fuse OUT	Row + WAGO 1	S→W	1.5 mm² red, soldered to the holder pins	Never through perfboard copper
P8	STAR− b2, b3, b4	Row − WAGO 1	W→W	1.5 mm² black, ferrule
P9	Row + WAGO 2–4	R1 +VM of BOB 1–3	W→S	0.5 mm² red, soldered at the BOB	9 wires in all
P10	Row − WAGO 2–4	R2 GND of BOB 1–3	W→S	0.5 mm² black, soldered	9 wires in all
P11	100 µF / 50 V	R1 (+) and R2 (−) of each BOB	·→S	On the pins, shortest legs	Stripe to R2, with P9 and P10
E-stop, 24 V, fans
X0	24 V adapter + / −	24 V+ / 24 V− WAGO	T→W	0.5 mm², barrel-jack terminal
X1	24 V+ WAGO	E-stop NC1	W→T	E-stop cable, core 1	4 × 22 AWG cable to the E-stop box
X2	E-stop NC1, other side	K1 coil + (terminal 1)	T→R	E-stop cable, core 2; red 6.3 mm receptacle	The coil has no polarity: call terminal 1 the + side
X3	K1 coil − (terminal 0)	24 V− WAGO	R→W	0.5 mm², red 6.3 mm receptacle
XD	1N4007 band on coil + (1)	anode on coil − (0)	R→R	Leads crimped in with the coil wires	Coil + is always +24 V
X4	E-stop NC2	Carrier X4: PA0 (CN7-28) and GND	T→T	E-stop cable, cores 3 + 4	Dry contact. 10k PA0 to 3V3 is on the carrier
F1	24 V+ WAGO	Fan red, × 3	W→·	Fan lead	One Orion OD4010-24HB per row, only above 1.4 A
F2	24 V− WAGO	Fan black, × 3	W→·	Fan lead	Never on the 30 V line (fan max 26.4 V)
U1	Nucleo CN1 (Micro-B)	Mac	U→U	USB-C to Micro-B	Powers the Nucleo and all 3.3 V logic

Logic: ribbon S1–S16

Ribbon conductor n = IDC pin n at both ends (red stripe = pin 1). On the carrier each Row column is one wire: morpho pin to IDC pin. All buses at 170 MHz / 32 = 5.31 MHz.

#	Signal	Row A · SPI1	Row B · SPI2	Row C · SPI3	Row board → BOB
S1	SCK	PA5 · CN10-11	PB13 · CN10-30	PC10 · CN7-1	L4 of all three
S3	MISO	PA6 · CN10-13	PB14 · CN10-28	PC11 · CN7-2	L6 of all three
S5	MOSI	PA7 · CN10-15	PB15 · CN10-26	PC12 · CN7-3	L5 of all three
S7	CS1	PB6 · CN10-17	PB12 · CN10-16	PD2 · CN7-4	BOB 1 L3
S8	CS2	PB10 · CN10-25	PB11 · CN10-18	PA15 · CN7-17	BOB 2 L3
S9	CS3	PB4 · CN10-27	PB2 · CN10-22	PB1 · CN10-24	BOB 3 L3
S10	EN1 (DRV_ENn)	PA9 · CN10-21	PB0 · CN7-34	PC7 · CN10-19	BOB 1 L9
S11	EN2	PC2 · CN7-35	PB7 · CN7-21	PC8 · CN10-2	BOB 2 L9
S12	EN3	PC3 · CN7-37	PC6 · CN10-4	PC9 · CN10-1	BOB 3 L9
S13	SLEEPn	PA8 · CN10-23, to all three headers			L10 of all three
S14	DIAG0	PA10 · CN10-33, to all three headers (optional)			J203-1 of all three (optional)
S15	3V3	CN7-16, to all three headers			L1 + ENC 3V3 of all three
S2 S4 S6 S16	GND	CN7-8 to IDC C, CN10-9 to IDC A, CN10-20 to IDC B (all common on the Nucleo); pins 2-4-6 bridged, 16 wired, on every header			L2 + ENC GND of all three
—	E-stop	PA0 · CN7-28 to X4, 10k to 3V3; X4 second pole to GND (CN7-20)			—

A1 keeps the bench's pins. Keep free: PA2/PA3 (VCP), PA11/PA12 (FDCAN), PA13/PA14 (SWD), PB3, PB8, PC4/PC5, PC13.

#	From	To	How	Wire	Note
Row board → each BOB's left-header holes (× 9), soldered at both ends
D1	Lead L1 (3V3)	L1 VCC_IO	S→S	F-F jumper half, housing cut off: soldered in the row board and in the BOB hole
D2	Lead L2 (GND)	L2 GND	S→S
D3	Lead L3 (CS1/2/3)	L3 CSN	S→S		Idle high
D4	Lead L4 (SCK)	L4 SCK	S→S
D5	Lead L5 (MOSI)	L5 MOSI	S→S
D6	Lead L6 (MISO)	L6 MISO	S→S
D7	Lead L9 (EN1/2/3)	L9 DRV_ENn	S→S		BOB pull-up: off until driven low
D8	Lead L10 (SLEEPn)	L10 SLEEPn	S→S		Held high
D9	Lead DG (DIAG0)	J203-1 DIAG0	S→S		Optional; open drain, wire-OR
—	L8 CLK	L7 GND	S→S	Bridge wire (a cut capacitor leg), bottom side	Internal clock, on every BOB
Motor (× 9)
M1	R4 A1	Motor black	S→·	Motor lead shortened, soldered into the BOB hole, heat-shrink	Coil A
M2	R3 A2	Motor green	S→·		Coil A
M3	R6 B1	Motor red	S→·		Coil B. Wrong direction: `invert_dir_mask`
M4	R5 B2	Motor blue	S→·		Coil B. Wrong direction: `invert_dir_mask`
Encoder, TBMA732 (× 9): wires on the connector legs, beep check as in step L
E1	CN1.1 VDD (leg)	Row board ENC 3V3 of its BOB	S→D	F-F jumper half (300 mm): cut end on the leg, female end on the row board's ENC pin	Never 5 V (MA732 max 4.6 V)
E2	CN1.2 GND (leg)	Row board ENC GND	S→D		Same nets as L1 / L2
E3	CN1.8 TEST (leg)	CN1.2 GND (leg)	S→S	Short insulated bridge	Outside legs 4 and 6 (CS, MISO)
E4	CN2.5 A (leg)	BOB R10 ENCA	S→S	F-F jumper half, housing cut off: soldered on the leg and in the BOB hole; the four twisted together, away from the motor leads	4,096 counts/rev out of reset
E5	CN2.4 B (leg)	BOB R11 ENCB	S→S
E6	CN2.6 Z (leg)	BOB R12 ENCN	S→S		Index
E7	CN1.8 leg (GND through E3)	BOB R7 GND	S→S		Return along A / B / Z

Power, E-stop and why

Supply	30 V / ≥ 10 A lab supply (borrow): limit ≈ 8 A, OVP 35.0 V. ≈ 1.8 A at rest at 1.4 A run current, up to ≈ 7 A with all nine moving (estimate).
E-stop	NC1 feeds the relay coil: pressed or a broken wire drops K1 and cuts 30 V to all nine BOBs. NC2 pulls PA0 to GND: the firmware latches and drives every DRV_ENn high. Releasing restarts nothing.
Why wired	9 pinch points at 48 mm pitch, up to ≈ 77 N per column at 1.4 A (estimate); a borrowed supply's button may be out of reach.
Logic	Nucleo 3V3 over USB, ≈ 210 mA for 9 encoders, 9 × VCC_IO and the MCU (estimate).
Cooling	Watterott heatsink on each BOB, as in step R1. Fans only above 1.4 A.
Ribbon	The pre-made 16-way cables in hand are 457 mm. Fold the extra length flat. SCK, MISO and MOSI each run next to a ground.
Relay	Omron lists only AC ratings for the G7L (25 A at 220 V AC). Both contacts in series share the 30 V DC break.
Encoders	The boards are 76 mm long: the 8 outer ones point outward, the centre one ≈ 27° off a row axis (≈ 4 mm clearance, estimate; `arm_deg` 27 for its tray). Joerg checks it on the frame.

Build the perfboards

Two designs: the carrier on the Nucleo, and the row board, built three times. Chip Quik plated boards, true 2.54 mm grid. Holes are named by row letter and column number, e.g. F12. On the carrier the two female headers hang underneath, so their pins are soldered on top; every other joint on both boards is underneath. Wire: 24 AWG solid, insulated. Leads to the BOBs: the 300 mm F-F jumpers cut in half, housing cut off, soldered at both ends (the BOB has bare holes). No crimping.

One board fits CN7 and CN10. UM2505 Fig. 6 puts the outer pin columns of CN7 and CN10 63.50 mm apart (= 25 × 2.54 mm) with the rows level. Step PB3 still dry-fits it before any solder.

Each drawing opens on group 1. Wire it, beep it, then Next. "Top side" shows what you see from above; "Underside" is the board turned over.

filled pad = a pin or wire end is soldered here empty ring = unused hole orange ring = hop hole big dot = wires join here