dial-me-spinner: mechanical design (rev A)

This document covers the mechanical design of the ~70 mm BLE dial spinner, which is used as a game/decision wheel, a Surface-Dial-style BLE controller and a fidget toy. It covers the enclosure, flywheel, bearing and axial clamp, magnetic detent, press-click suspension, antenna placement, assembly and mechanical BOM.

• Architecture of record: hardware/design/SPEC.md §8. SPEC wins on any disagreement. Section 3 lists where SPEC cannot be built as written, what the model does instead (tagged DEV: in the model) and what SPEC should change to. It also records the PCB placement problems the model found and how they were fixed.
• Model: hardware/mechanical/dial_spinner.scad is a parametric OpenSCAD 2021.01 model of the whole device. Every SPEC number is a named constant.
• Board parts come from the real PCBs. extract_board_parts.py reads hardware/kicad/{main,base}/*.kicad_pcb (KiCad 9 pcbnew) and writes board_parts.scad/.json: one oriented box per footprint (Fab outline × datasheet height), plus the mated plugs of the side-entry connectors, the module's mated u.FL and a solder-joint body on the NFC pads J11. So every clash check below tests the boards as they are placed now.
• Calculations: hardware/mechanical/calcs.py prints everything quoted here; the full output is in calcs_output.txt. Its section 11 reads the model's geometry directly (OpenSCAD view="prims" + board_parts.json).
• Interference check: check_clashes.sh runs CGAL booleans: rotor envelope, press travel, rest, and same-body pairs (board parts vs actuator, hub, harness, coax; battery leads vs cup, plate and guide pins; NFC coil and tail vs the board, its top parts, the OLED bracket, glass, light box and cover). Result: clash_report.txt.
• Parts: hardware/mechanical/stl/ (binary STL, oriented for the bed). render_all.sh regenerates every render, STL and report.

Anything marked (to verify on prototype) is an assumption or a model result that has not been measured yet. Section 15 collects them into a checklist.

---

1. Concept

Body	What it is	How it moves
Base	PA12 cup (Ø70, z 1.0–14.3) on 3 micro-suction arc pads. It holds the LiPo (long axis θ 0/180°), the base board (USB-C, charger, press switch), the 2.4 GHz flex antenna on a printed shelf at θ 90°, and 3 brass guide pins with stop screws.	Sits on the desk
Floating assembly (FA)	3-spoke spider plate + turned 316L spindle + brass spacers + main board (MCU, OLED, LED ring, TMAG5273, detent actuator) + optics + top cover (with the NFC tag coil). One M4 nut on the board top clamps the whole stack. It is SPEC's "stationary hub".	Does not rotate. Three springs push it up against three stop screws. A press moves it down 0.45 mm to the click and at most 0.50 mm to the hard stop.
Rotor	316L flywheel ring (OD 70 / ID 54 / 10, turning only) with 48 N52 Ø2×2 magnets at R 29.0 carried in a printed shim that drops into a turned recess in the ring's top face, bonded to a PA12-GF carrier that holds the 608 outer race on a compliant seat	Spins on the 608 around the spindle and moves down with the FA when pressed

The magnet track does three jobs: detent (a stationary Ø3×2 N52 magnet 1.0 mm above the track gives 24 detents of 15° and slides radially inward along θ = 90° to weaken it, free spin at 7.5 mm), encoder (TMAG5273 on the main-board underside at r 29, θ 270°, 24 electrical periods per revolution) and wake source. The 12-way link and the 1.13 mm antenna coax pass through the hollow spindle, so nothing rotates in the electrical path.

Assembly (USB-C at θ = 0°, front right)	Exploded view, in assembly order (§12)
Main-board underside from below (parts from the PCB). Rotor envelope ghosted; SPEC §8 bottom rules outlined: keep-out θ 60–120° (yellow), actuator corridor θ 30–160° (orange), rotating hub zone r 9.5–12.5 (cyan); harness red, coax grey.	Base from above, FA plate ghosted: antenna shelf (top), J3 and J2 with their mated plugs, guide pins

Key numbers (from calcs.py):

Quantity	Value
Overall	Ø70 × 36 mm, ≈ 227 g (rotor 134 g, rest of FA 50 g, base 43 g), CoG z ≈ 19.4 mm
Rotor	134 g, I = 1.19·10⁻⁴ kg·m² (the ring is 96 % of I), CoG 0.8 mm below the ball plane
Energy	59 / 235 / 529 mJ at 5 / 10 / 15 rev/s
Free spin from a 10 rev/s flick	95–131 s (SKF model, ν 20–10 mm²/s); 35 s if friction is 2.7× the model. Acceptance gate ≥ 35 s, target ≥ 45 s (§4).
Detent	24 × 15°, 16.5 mN·m at full click, continuously variable to < 3 µN·m (free)
Click force	4.6 N at the centre, 1.9–2.4 N at the ring rim (varies with azimuth)

---

2. Stack-up

z is measured from the desk. θ = 0° is the USB-C direction, CCW seen from above. "Model" is what dial_spinner.scad builds. ⚠ marks a deviation that section 3 explains.

z (mm)	Part	SPEC §8	Model
0.0–1.0	feet	3× at r 28	⚠ 3 arc pads, 70° each, r 25–31, at θ 0/120/240°
1.0–2.5	base cup floor, PA12	OD 70, wall 1.6, wall top 14.3	as SPEC; bottom chamfer 0.6
2.5–8.5	LiPo bay	pack ≤ 6.0 thick, |x| ≤ 27.5, |y| ≤ 17.25, leads at θ 180	as SPEC (envelope 55 × 34.5 × 6.0). BOM pack 503450-class, ≤ 5.6 max (§13). Leads (28 AWG SH, two per rail) leave the θ 180 end at y ≈ −7.5, clear of the θ 180 guide column, run beside the pack at z ≈ 10 under the spider and come up over the base-board edge at θ ≈ 222° to J2.
2.5–5.5	antenna shelf (printed, part of the cup)	"printed shelf, centred on the chord y = +23.5 (strip y 19.0–28.0)"	shelf y 18.8–33.4, |x| ≤ 18.25
5.5–5.75	antenna Molex 146153-0100 (34.9 × 9)	centred on y = 23.5, corners ≤ r 33, ≥ 1.75 mm from the pack	as SPEC (corners at r 33.0)
2.5–9.5	base-board supports	one M2 boss at r 20, θ 270; rib under the θ 90 edge (y 17.75–18.75); USB-C shell	as SPEC: boss with an M2 × 3 heat-set insert (0.25 mm from the pack envelope); rib |x| ≤ 6, no screw; USB shell captured in a wall slot
9.5–11.1	base board Ø46 + USB tab to r 34.6	4-layer 1.6 mm, parts on the top only	from the PCB
11.1–12.6	TS-1187A at the centre		body 5.1 × 5.1 × 1.0 (verify), actuator Ø2.5 to 12.6
11.1–14.36	base-board parts	tall parts in the open sectors	from the PCB; tallest: USB-C J1 (top z 14.36), side-entry SH J2 and J3 (2.95 tall, top z 14.05)
2.5/6.5–11.5	3 guide columns / ear	pins + springs at r 29, θ 60/180/300	θ 300: column from the floor. θ 180: ⚠ column from the floor at r 30. θ 60: ear z 6.5–11.5 above the antenna. Stop face (top) at 11.5. Ø3 brass pins pressed + bonded + Ø1 cross pin.
12.0–14.3	FA spider plate PA12-GF	3 spokes ±18° at 60/180/300, hub r ≤ 6, outer ring r 26–30 with a USB notch (|y| ≤ 5.6), pre-gap 0.15	as SPEC; under-bosses Ø6.4 at the guides (z 12.0) are the hard stop: 0.50 mm travel
13.5–14.8	M2 × 3 brass stop screws	"preload ≈ 4 N up against stops"	heads in Ø4.6 counterbores (floor 13.5)
12.75	press pad face	"centre boss over the switch, pre-gap 0.15"	Ø3 × 1.0 silicone pad in the spindle bottom; pre-gap set per unit (0.03–0.23, §8)
12.8–29.05 (+ spigot to 32.3)	spindle, turned 316L, Ø8 h6, bore Ø6	hollow Ø8 × 1.0, z 12.8–29.1, clamped to plate and board, slots at z ≈ 14.5–16 and under the board	bottom flange Ø10 × 0.6 in a D-recess under the plate; solid end with the pad pocket; slots θ 120 (z 14.45–16.35) and θ 270 (z 25.8–28.25), 5.0 wide; ⚠ M4×0.5 spigot through the board's Ø5 hole
14.3–18.0	brass lower spacer, Ø11.5, with drop-stop flange Ø21.2 (z 16.40–16.85) and a wire window at θ 120		0.15 mm under the rotating carrier lip
15.6–25.6	rotor ring 316L	OD 70 / ID 54, track on the top face, R_t 29.0	chamfers 1.0 top-outer, 1.5 bottom-outer (USB plug clearance), 0.5 inner. Straight knurl z 17.1–24.1. 4× M2 balance holes in the bottom face at r 31.
17.0–25.5	rotor carrier PA12-GF, post-machined	web z 17–19, r 11–27	web r 12.2–26.95 (bonded into the ring ID over z 17–19); collar r 11.0–12.2 bored Ø22 H7; lip r 9.9–11 under the race. No ribs.
18.0–25.0	608-2Z	outer race r 11	slip fit in the H7 collar on the lip; wave washer z 25.0–25.5; 0.5 mm 316 cap ring bonded on the collar top, z 25.5–26.0
25.0–29.1	brass hub: inner-race spacer + flange r ≤ 8.5 with 3× M2 tapped at r 6.5 (θ 90/210/330), C-gap at θ 270 for the wires	"3× M2 at r 6.5 (spindle flange)"; rotating hub top z 26.0	the rotor hub (cap ring) tops out at 26.0 (§3 A6, now in SPEC). The main-board bottom under the clamped flange (r < 8.8) carries no tracks or vias, only the GND pads H1–H3 and the GND pour (the hub is GND).
19.8–29.1	bowl: stepper (Ø8 can at x 6.5–15.5, y 15.5, axis z 23.8), saddle to the θ 60 board hole	"chord placement near θ = 90°"	sweeps θ ≈ 35–160° (§7)
26.6–29.08	detent carriage: Ø3×2 N52 (z 26.6–28.6), 0.48 plate between two rails on the board bottom, 0.8 mm FR4 link plate (z 26.7–27.5) with the inclined slot	1.0 above the track, r 29 → 21.5 along θ 90; carriage plate 0.5	as SPEC, except the plate is 0.48 in a 0.50 gap (§7)
27.65–29.1	TMAG5273 (bottom), package at r 29.4, θ 270°	Hall element at r 29, θ 270	Hall element 3.13 mm above the track face
29.1–30.7	main board Ø64, 4-layer 1.6 mm, parts on both sides	centre Ø5 + 3× plated GND M2 r 6.5 (H1–H3) + 2× NPTH M2 r 17 θ 60/120 (H4/H5)	holes from the PCB; the board sits on the hub, clamped by the M4 nut
30.7–32.2	M4×0.5 brass nut, OD 7	top r ≤ 4.6 kept free for it	on the board top under the OLED, in the bracket's centre hole
30.7–31.6	48 × WS2812C-2020 at r 28.5		from the PCB
31.7–34.2	light box r 27–30, walls 0.6, 24 septa	31.6–34.2	walls stop 0.1 above the LED tops
31.8–32.8	OLED bracket: rim under the glass edge + 3 spokes to the hub-screw bosses		pocketed 0.3 mm round / 0.2 mm above every top part taller than 0.95 mm; relief over J11 and the NFC tail (rim underside raised to z 32.3, 0.5 mm rim left, §9)
33.0–34.7	OLED glass 33.8 × 34.0	active-area centre on axis	glass centre at y −1.47; 0.2 mm foam tape
34.2–36.0	diffuser r 27–30 + top cover r ≤ 32.5, Ø30 window	NFC flex coil in the cover, tail soldered to J11 on the main-board top (r 21.5, θ 222°)	0.7 mm chemically strengthened cover glass Ø33 bonded in the cover (z 35.3–36.0); 0.6 mm air gap to the OLED glass. Underside raised to z 35.0 inside r 26.35 for the NFC coil (replaces the rectangular glass pocket); a 0.6 mm wall ring r 26.35–26.95 carries the 6 keys.
34.80–34.95	NFC coil: annular flex Ø31/Ø52.5 × 0.15, copper turns r 24.9–26.15	"the 13.56 MHz NFC flex coil sits in the cover"	bonded (0.05 PSA) to the cover ceiling, 0.1 above the OLED glass; tail = 2.0 mm tongue folded down at θ 222°, r 24.1, onto J11 (§9). Dimensions TVP.

Running clearances (nominal, calcs.py §11).

Pair	At rest	Pressed (0.50)
Ring bottom to cup wall top	1.30	0.80
Ring bottom to stop-screw heads (base-fixed)	0.80	0.30
FA plate / spokes to base-board top (parts ≤ 1.0 tall allowed under them)	1.70	1.20
Plate top to carrier web (wires and coax run here)	2.70	2.70
Detent magnet to track	1.00	1.00 (both on the FA)
Carrier lip to brass drop stop	0.15	0.15 (intended crash stop)
Stepper bracket bar to rotating web	0.80	0.80
Stepper front plate to collar (plan)	0.55	0.55
Rotor hub top (cap ring, 26.0) to board bottom	3.10	3.10

Interference check (clash_report.txt, CGAL, plus the per-part tables in calcs_output.txt §11). With the current boards everything clears except the intended contacts listed in clash_legend.txt: the plate under-bosses on the cup stop faces at full travel (hard stop), the SW1 actuator against the press pad, the bonded detent rails and saddle on the board bottom, and two near misses at 0.3 mm margin (the felt shoe resting on the ring, and the drop-stop flange 0.15 mm under the carrier lip). The NFC coil and its tail are in the check too, and all their pairs are clear. The PCB clashes found earlier are fixed (§3B). The tightest remaining gaps:

• Antenna coax between U12 and the module's inner corner: 0.84 mm to U12, 0.64 mm to C10, 0.25 mm to U10 and 0.44 mm above the rotating cap ring (r 9–13). It must be tacked up with Kapton tape at assembly (§12 step 12.4).
• Harness under C11 (0402 at r 9–10): 0.73 mm.
• C14 (crystal load cap, r 9.0–9.7) to the brass hub: 0.54 mm.
• Coax over base J3 and its plug: 0.47 mm at rest, 0.22 mm pressed (the free span sags with the FA).
• J40 (motor connector, 2.95 tall): 0.35 mm to the slider nut at click, 0.55 mm to the rotor envelope.
• U10 MDBT50Q to the rotor envelope: 0.90 mm; to the brass hub 1.05 mm.
• USB-C J1 (base) to the ring: 1.24 mm at rest, 0.74 mm pressed.
• Battery leads to the pressed spider plate: 1.40 mm (r 30.2, θ 196°, z 10).
• NFC tail beside the OLED glass edge: 0.34 mm, and 0.54 mm to the bracket rim. All three ride on the FA, and the flex may touch the glass without harm. The J11 joint has 1.25 mm under the rim relief (0.75 mm without it).

Board underside from below: red = interference with the rotor envelope (ghosted), orange = FA-internal interference (board parts vs actuator, hub, harness, coax). Regenerate with render_all.sh after any placement change.

---

3. Where SPEC §8 cannot be built as written, and what the PCBs had to change

3A. SPEC items (for the SPEC owner)

#	SPEC says	Problem (numbers)	Model / recommendation	Severity
A1	(adopted in SPEC rev of 02:50) Antenna on chord y = +26, ≥ 4 mm from the pack	At y = 26 the corners are at r 35.1, outside the cup ID (33.4). With corners at r ≤ 33 the centre can be at most y = 23.5, which leaves 1.75 mm to the pack.	SPEC now: y 23.5, ≥ 1.75 mm, base-board copper keep-out. Remaining RF risk: a ≤ 30 mm-wide pack (|y| ≤ 15) would restore 4.0 mm (~10–20 % less capacity); otherwise measure S11 (§15 #9).	Medium (RF)
A2	(adopted) Base board: 2× M2 bosses at r 21, θ 90/270	A boss at (0, 21) sat on the antenna. Moving it to θ 60/120 at r 21 does not help: a Ø5 boss there reaches y 15.7, inside the pack.	SPEC now: one boss at r 20, θ 270 + a rib under the θ 90 edge + the USB shell (captured in a wall slot: the board slides in radially, then the θ 270 screw locks it). Note: at r 20 the Ø5 boss is only 0.25 mm from the pack envelope; r 21 gives 1.25 mm.	Low
A3	Guide pins + springs at r 29, θ 60/180/300	θ 60 is above the antenna, so the pin cannot reach the floor. θ 180 at r 29: the pin (x −30.5…−27.5) touches the pack end (x −27.5), which is also where the leads exit.	θ 300: column from the floor. θ 180 moved to r 30 (column from the floor at x ≤ −27.8, 0.3 mm from the pack). θ 60: an ear z 6.5–11.5, 0.75 mm above the antenna. Leads must leave the pack end at |y| ≥ 4 (BOM). SPEC fix: θ 180 guide at r 30, or a pack ≤ 53 mm long (|x| ≤ 26.5).	Medium
A4	Guide hardware at θ 60	A brass pin, steel spring (68 mm of wire, ≈ λ/2 at 2.44 GHz) and screw sit 0.75–9 mm above the antenna's +x end.	Fit PEEK pin, PEEK cross pin and nylon screw at θ 60 from the first build. Keep the 302 spring unless S11 says otherwise (§9). The pull-out is lower with PEEK (§8).	Medium (RF)
A5	608 with "≥ 2.5–3 N wave-spring axial preload"	A spring between the outer race and the rotor cannot preload a single bearing, because both rotate. The only axial load is gravity (1.31 N) ± detent forces. The click-mode detent tilt moment ≈ 0.7 N × 29 mm ≈ 20 mN·m, against a gravity moment capacity Fa·dm/2 ≈ 10 mN·m.	The wave washer is a compliant drop seat (§10). For real preload: (a) gravity only (accept rocking; measure), (b) axisymmetric magnet-ring pair (rotor ring on the carrier underside + stator ring on the FA plate, ≈ +1.5–2 N, no cogging if uniform), (c) duplex 608 (≈ 2× friction, +7 mm). Fa 2.7 N costs 95 → 62 s (§4).	Medium
A6	(adopted) "rotor hub tops out at z ≈ 25"; harness lies flat under the board on that basis	z 25.0 is the 608's own top face. Any drop-compliant outer-race seat (verifier requirement) needs a spring and a retainer above it. The minimum is a 0.5 mm wave washer + 0.5 mm cap: top 26.0.	SPEC now: hub top 26.0, and bottom parts over r 9.5–12.5 keep their bottom at z ≥ 26.3 (hang ≤ 2.8 mm). The current main board meets it (calcs_output.txt §11: no part flagged). In the model the harness and coax cross the hub zone with their centrelines at z ≥ 26.9; the coax, at the U12 / module corner, is the closest (0.44 mm above the cap ring, Kapton-tacked, §2).	–
A7	Spindle Ø8 × 1.0 tube z 12.8–29.1 "clamped to the lower plate and to the main board"	A plain tube cannot be clamped at both ends. The bearing must go on from one end and something removable must hold the other. The old model (bonded socket, sliding hub) held nothing axially (review).	Turned 316L spindle: bottom flange in a D-recess under the plate, solid bottom with the press pad, M4×0.5 spigot through the board's Ø5 centre hole + brass nut on the board top (0.10 N·m ≈ 120 N clamp). One nut clamps board → hub → inner race → spacer → plate → flange (§4). The PCB rules are in SPEC and layout.py: main-board top r ≤ 4.6 kept free for the nut (OD 7: no parts, tracks or vias), no tracks or vias on the bottom under the clamped hub flange (r < 8.8) and none within 0.35 mm of the Ø5 hole on any layer (r < 2.85). SPEC §8 stack-up now describes the turned spindle, its M4 spigot and nut, and the brass hub.	High
A8	(adopted) FA plate outer ring r 26–30 (continuous)	The USB-C receptacle (z up to 14.36) sits at r 27.7–35 under the ring.	SPEC now has the notch |y| ≤ 5.6.	–
A9	Pre-gap 0.15, "FA moves 0.3–0.5"	TS-1187A travel is 0.25 ± 0.1 (verify). A fixed pre-gap and hard stop cannot cover that spread without crushing or missing the switch (the old 0.5 mm TPU dot put 14–26 N through the 1.6 N switch).	Hard stop 0.50 (SPEC's pressed clearance 1.2 mm kept). Silicone pad k ≈ 20–25 N/mm. Pre-gap set per unit = 0.38 − measured travel (0.03–0.23). The switch then sees 2.7 N (≤ 3.9 N worst) at the stop. Reject switches with travel > 0.32. §8.	Medium
A10	3 feet at r 28	A rim press tips the 227 g device at 1.60 N, below the rim click force (1.9–2.4 N).	3 × 70° arc pads (r 25–31): tip edge 28.1 mm, tip force 11.6 N.	High (Surface-Dial use)
A11	(adopted) Actuator in the bowl, chord near θ 90; bottom keep-out θ 60–120	A ~33 mm VSM08102 cannot lie radially (hub r ≤ 12.2, ring r ≥ 27). The working chord (y 15.5, inclined-slot link) sweeps θ ≈ 35–160°.	SPEC now: bottom-side parts in θ 30–160°, r 8.5–27 must keep their bottom at z ≥ 27.8 (≤ 1.3 mm tall), except where calcs_output.txt §11 shows clearance. Two taller parts reach into it: J40 (2.95 mm), which clears the slider nut by 0.35 mm and must not move toward the corridor, and the piezo BZ1 (1.9 mm, r 22, θ 22°), whose edge reaches θ ≈ 36° and clears the actuator by ≥ 1.0 mm (not listed in §11). The θ 120 board hole is unused.	–
A12	SPEC §7: no ferromagnetic part within 10 mm of the track; SPEC §8: an 8 mm stepper in the bowl r 12–27	Conflict. Any Ø8 can in the bowl is within 10 mm of the track wherever r > 17. The VSM08102 can comes within 3.1 mm of the nearest track magnet (r 24.9 vs magnet edge r 28, overlapping z), and its steel lead screw within 4.6 mm. The can cannot move inboard: the collar (r 12.2, top 26.0) blocks it.	Rough estimate of the can's cogging on the alternating track: 10–100 µN·m, the same order as the bearing drag at 0.1–1 rev/s (18–44 µN·m). It may give preferred rest angles in FREE mode. Bench test before the PCB order (§15 #2). Non-ferrous options: piezo (e.g. squiggle-type) or SMA actuators. Confirm the MDBT50Q shield-can material too.	High (to verify)
A13	Light box z 31.6–34.2, r 27–30	The LED research wants a 4.0–4.5 mm air gap; this gives 2.6 mm.	Walls 0.1 above the LED tops; 24 septa (one per detent): a segmented "prize wheel" look. Uniformity to verify.	Low
A14	Detent LUT (research) at R_t 30 on the ring underside	SPEC: R_t 29 on the top face (pitch 3.80 vs 3.93 mm)	Expect ~2–5 % lower torque. Firmware self-calibrates from the in-well f_n.	Low
A15	(fixed in SPEC) §2 said base board "Ø40", and "battery 0.5 mm below"		SPEC now Ø46 and 1 mm	–

3B. PCB placement findings (for the layout owner; from board_parts.scad and the clash check)

All closed. The board parts were re-extracted from the routed boards on 2026-10-19, after the last placement changes (among them U11, U12 and Y1 on the bottom and J10, TP10–TP13, the VSYS_SENSE divider and the OLED IREF/VSL parts on the top moved beside the pins they serve; D20 is now under the OLED bracket rim, which is pocketed over it), and check_clashes.sh and calcs.py pass on them: intended contacts only, no part needing action (§2). Re-run extract_board_parts.py, check_clashes.sh and calcs.py after every placement change. Fixed earlier: J1 and J3 face the centre, J2 sits at θ 222° with its mouth outward, Q1 left the spoke zone, RT1 moved to the top side, the θ 90 base-board hole is gone and the antenna copper keep-out is in layout.py. The clamp's copper keep-outs are in layout.py too (A7): no tracks or vias on the bottom under the hub flange (r < 8.8), on the top under the M4 nut (r < 4.6) or within 0.35 mm of the centre hole.

#	Part as first placed	Problem	Done
B1	U10 MDBT50Q at r 14.5, θ 215, u.FL end inward (inner end at r 6.8)	The module overlapped the brass hub (r ≤ 8.5) by 1.5 mm (static clash). The mated u.FL (bottom z 25.8, at r 6.7–10.3) clashed with the hub and sat 0.2 mm into the rotating cap ring (needs ≥ 0.3). The coax could not reach it except through the hub.	Closed. U10 is centred at r 17.3, θ 215° on the bottom, inner end r ≥ 8.8, u.FL end facing outward (mated u.FL at r ≈ 23). Clearances: rotor 0.90 mm, hub 1.05 mm. The coax runs back under the board beside the module to the θ 270 top slot, threading between U12 and the module's inner corner (0.84 mm to U12, 0.25 mm to U10, 0.44 mm above the cap ring): Kapton-tacked at assembly. Modelled coax path 79 mm.
B2	J1 (SM12B, 2.95 tall) at r 17.5, θ 330, mouth toward the centre	The back of the mated plug reached r 11.9 at z 26.15, 0.15 mm above the rotating hub (needs ≥ 0.3).	Closed. J1 moved out to r 19.5 (θ 330°, bottom, entry facing the centre). J1 and its plug clear the hub, the rotor and the wires by ≥ 1.0 mm (not listed in calcs_output.txt §11). Harness path 43 mm.
B3	J2 (S2B-PH-SM4-TB, 5.5 mm tall), base r 17, θ 222, mouth outward	Top z 16.6 under the rotor web (17.0 at rest, 16.5 pressed → −0.10). The mated plug reached r 26.3, into the FA ring (z 12.8–14.3), and the battery leads crossed the ring.	Closed. J2 is a side-entry JST SH SM04B-SRSS-TB (LCSC C160404, the motor connector's part), 2.95 tall, pins 1–2 BAT+, 3–4 GND, at r 17, θ 222°, mouth radially outward (SPEC §6/§8). The pack is re-terminated with 28 AWG pre-crimped SH leads, two per rail; they run beside the pack at z ≈ 10 under the spider and come up over the board edge at θ ≈ 222°. J2 and its plug clear the rotor and plate by ≥ 1.5 mm; the leads clear the pressed plate by 1.40 mm.
B5	R42 (0402, LED_D48 pull-down), main top, r 30.4–31.05	Under the top-cover skirt (r ≥ 30.6)	Closed. R42 is at r 24.1, inside the r 25.3 top-side limit.

---

4. Bearing, axial clamp and lubrication

Choice: SKF 608-2Z (8 × 22 × 7, C 3.45 kN, C₀ 1.37 kN, 13 g; datasheet). Equivalents: NSK/NMB 608ZZ. SKF 608-2RSL is acceptable: SKF's model gives zero seal friction for RSL at D ≤ 25 mm, and it keeps oil better. With 2RSL, turn the hub and spacer clamp faces to Ø10.0–10.5. Never use RSH or RS1 contact seals (≈ 7.6 N·mm: the rotor stops in about 1 s). Buy 5+, measure each, keep the best two.

Spin-down with the rotor computed here (Fa = 1.24 N from gravity, plus eddy and air terms of < 3 %), from the SKF frictional-moment model (Table 2a constants; it reproduces the verifier's 0.147 N·mm for the 608 and 1.6 N·mm for the 61808):

Bearing / lubricant	from 5 rev/s	from 10 rev/s	from 15 rev/s
608-2Z, oil ν = 10 mm²/s	87 s	129 s (475 rev)	159 s
608-2Z, oil ν = 20	64 s	93 s (332 rev)	114 s
608-2Z, ν 20, friction × 2.7 (planning floor)	24 s	35 s	43 s
608-2Z, factory grease after run-in (ν 200)	20 s	28 s	34 s
61808 (6808), oil ν 10 / ν 20	10 / 7 s	14 / 10 s	17 / 12 s

Uncertainty and acceptance. The model runs below the bearing's minimum load and with drop lubrication. The one measured comparison (verifier) suggests it under-predicts friction by up to 2.7×. Therefore:

• Acceptance gate: ≥ 35 s from 10 rev/s. This equals the 2.7× planning floor, so a unit that performs as the pessimistic model predicts passes. Below the gate the bearing prep failed (residual grease, dirt, shield rub): re-clean or swap the 608.
• Target: ≥ 45 s (the verifier's 45–60 s band). Record the measured value: it calibrates the friction model.
• Expectation if the model holds: 93–129 s.

The magnet shim (§5) costs 5.3 % of the rotor's inertia, which moves the ×2.7 floor from 35.3 s to 34.5 s. The gate stays at 35 s because the shift is 2 % inside a model whose stated uncertainty is 2.7×, but note that the baseline only ever cleared its own floor by 0.3 s: this gate has never had margin, and the first measured unit decides it.

The ≈ 10× advantage of the 608 over a 61808 holds either way, because the same model error applies to both.

Lubrication (every bearing, before assembly).

1. Soak the 608-2Z in IPA and run it 10 min in an ultrasonic bath. 2Z shields are pressed in, so leave them in place. Repeat with fresh IPA, then blow dry.
2. Add 1–2 drops of ISO VG 2–10 synthetic instrument/spindle oil through the shield gap (e.g. Mobil Velocite No. 3 = VG 2 or No. 6 = VG 10; base oil not stated, to verify).
3. Spin it by hand for 1 min, wipe the outside, and spin-test it on an Ø8 arbor: > 30 s from a finger flick, no shield rub, no gritty stop.
4. Maintenance: log the spin-down monthly with the encoder (firmware) and re-oil below 70 % of the as-built value.

Axial clamp (new: the old hub slid freely on the spindle). The stack is held by one removable nut:

M4x0.5 brass nut (board top, 0.10 N*m, Loctite 222)
  -> main board (Ø5 hole over the Ø4 spigot)          z 29.1-30.7
  -> brass hub (inner-race spacer, flange r <= 8.5)    z 25.0-29.1
  -> 608 inner race                                     z 18.0-25.0
  -> brass lower spacer (+ drop-stop flange)            z 14.3-18.0
  -> FA plate hub (PA12-GF, 0.9 mm in the clamp)        z 13.4-14.3
  -> spindle bottom flange (Ø10 x 0.6, D-recess)        z 12.8-13.4   <- the spindle is the tie rod

• Clamp: 0.10 N·m ≈ 120 N (K ≈ 0.2). That is 4.4 MPa on the plate hub. Thread root of the M4 spigot: 7.7 mm², ≈ 1.5 kN.
• Creep: the PA12-GF ring will creep a few µm and lose most of the preload over time. The joint still cannot separate, because the nut is thread-locked, and the residual play is µm-level. Retighten at service.
• Angles: the spindle has D-flats at θ 90 (flange, spacer and hub zones; the 608 seat z 18–25 stays round). The plate, spindle, spacer window, hub C-gap and board holes therefore all line up.
• Inner race: 608 on Ø8 h6, with Loctite 641 if the fit is loose. The inner race only ever sees the clamp, never a load through the balls.
• ESD path (SPEC §8): ring → conductive epoxy → 0.1 × 3 mm phosphor-bronze strap along the web and up the collar → 316 cap ring → wave washer → 608 outer race → balls/oil film → inner race → 316L spindle → brass hub → hub face on the plated H1–H3 pads → GND. It is metal all the way except the bearing's oil film, which bleeds static charge (MΩ). Pass: ring-to-GND < 10 MΩ with the rotor turning slowly (to verify).

Outer race seat (compliant, drop-protected, §10). The race slips into the carrier collar, which is bored to Ø22 H7 after printing (22.000–22.021; the race OD is 22 −0.009/0). It sits on a rigid lip below. Above it are a 22 mm-bore wave washer (working height 0.5, 30–50 N, ≤ 300 N/mm) and a 0.5 mm 316 cap ring (OD 24.4, ID 19.9) bonded with epoxy onto the collar top face (88 mm², ≈ 400 N). To replace the 608, warm the cap with an iron to ~120 °C and lift it off. The slip torque of the clamped race (≈ 40 N × 0.2 × 11 mm ≈ 90 mN·m) is far above the bearing drag (0.15 mN·m), so the race does not creep.

Axial preload (SPEC A5). Effect of the net axial load:

Fa (N)	M at 10 rev/s (N·mm)	Fa·dm/2 tilt capacity (mN·m)	Spin from 10 rev/s
1.31 (gravity)	0.147	9.9	95 s
2.0	0.179	15.0	75 s
2.7	0.207	20.2	62 s
3.5	0.236	26.2	53 s

Prototype with gravity only first. Measure click-mode rocking with the TMAG (a once-per-detent wobble in the Bz amplitude). Add a magnet-ring preload only if the rocking is audible or felt (to verify on prototype).

---

5. Flywheel: material, drawing notes and balancing

Material: 316L stainless. ρ 7.98 g/cm³, non-magnetic, corrosion-proof under fingers, σ ≈ 1.35 MS/m (about 11× below brass, so the stationary magnet's eddy drag stays at a few µN·m). Turning only — the 48 magnet pockets have moved into a printed shim, below. Net ring mass 108.3 g. Price: §13 item 7.

Feature	Spec
OD / ID / height	70.00 / 54.00 +0.02/0 / 10.00 mm
Concentricity OD↔ID	≤ 0.02 TIR
Faces	flat and square ≤ 0.02. The top face outside the recess is the detent/encoder datum and the damping-pad land (§7). Face runout ≤ 0.05 TIR (the detent loses ~10 % torque per +0.1 mm gap).
Chamfers	1.0 × 45° top-outer (flick edge); 1.5 × 45° bottom-outer (USB-C plug overmold clearance, §9); 0.5 × 45° bottom-inner. The top-inner chamfer is gone — it fell inside the recess
Knurl	straight, DIN 82 RAA 0.8, OD band z 17.1–24.1
Shim recess	Ø62.00 × 2.00 +0.02/0 deep in the top face. Its floor is the magnet seat, so flat and square ≤ 0.02: the magnets bottom on steel and it is this depth, not the printed shim, that sets the 1.0 mm detent gap
Balance holes	4 × M2 tapped, 3 deep, in the bottom face at r 31, θ 45/135/225/315°
Process order	rough-turn → knurl → N52 cling test → only if it clings: solution-anneal (1050 °C, water quench) → finish-turn faces, ID, chamfers and the recess to the TIR/flatness specs → drill/tap balance holes → repeat the cling test. Never anneal a finished ring (distortion, scale). Never nickel-plate (ferromagnetic).

Magnet shim (printed). A Ø62 / Ø54 × 1.95 ring with 48 Ø2.10 through holes at R 29.000, every 7.5°, that drops into the ring's recess and is located radially by it. It exists to get the 48 blind pockets out of the steel: they were a flat-bottom milling operation held to ±0.02 in stainless, and as through holes in plastic they need no depth control at all. The quote bears it out — §13 item 7 fell from $71.45 to $52.49, machining $60.01 → $41.05.

Three things follow, and the third is the open risk:

• The gap stack loses a term. The magnets bottom on the recess floor and stand 2.0 proud of it, so the track height is set by the ring and the magnets' own height. The shim is printed 1.95, deliberately 0.05 shy, so its printed thickness never enters the stack. The gap closes 0.05 mm versus the old pocket floor (+5 % detent torque).
• Assembly decouples. The 48 magnets are bonded into the shim in a flat jig, in alternating polarity, off the ring and testable before commitment. A mis-built track now scraps a $0.32 printed part instead of a $52 one.
• Angular accuracy is the risk. The track is a 24-pole-pair encoder, so a magnet 0.10 mm off tangentially is 0.198° mechanical = 4.74° electrical — against the 3.5° §14 already budgets for XZX sampling. The scatter is static, so the firmware's calibration spin can absorb it into a 48-entry LUT; detent feel cannot be calibrated, but 0.10 mm on a 3.79 mm pitch is 2.6 % and is a prototype question. If printing proves too loose, a laser-cut sheet shim holds ±0.05 mm for about the same money.

Walls are the binding constraint, not inertia. 0.95 mm between the Ø2.10 bores and both recess walls clears JLC3DP's 0.8 mm hard minimum but sits under its 1.2 mm recommended wall, so prototype before committing. This is what fixes the recess at Ø62: narrowing it to Ø61 would buy 1.8 % of inertia back and leave 0.45 mm walls, which is unmanufacturable. Widening past Ø63.2 would put the damping pad (r 31.6–32.4, §7) onto plastic.

Rotor carrier. PA12-GF (SLS) or MJF PA12-GB, 5.6 g, no ribs (the old ribs at z 16.2–17 sat below SPEC's web and in the wire space). Print with +0.4 mm stock on the bore and register, then bore Ø22 H7 and turn the register to Ø53.95 in one chucking (bore-to-register ≤ 0.02 TIR). No aluminium or copper may rotate near the detent magnet (verifier). The ESD strap is a 3 mm-wide radial strip, not a loop, so its eddy drag is negligible.

Alternatives. C360 brass ring (132 g; about 11× the eddy drag in click mode). RF-transparent ring (§9).

Balancing. Target ISO 1940 G2.5 at 15 rev/s: U_per = 3.36 g·mm for the 127 g rotor; aim ≤ 2 g·mm.

Error source	A: ring bonded freehand	B: bonded on an arbor (recommended)
Ring OD/ID eccentricity (0.02 TIR)	1.08 g·mm	1.08
Ring-to-bearing eccentricity	3.92 (e ≈ 0.035)	1.12 (TIR ≤ 0.02)
48 magnets, mass σ 3 %, random	0.28	0.28
5 mg glue blob at r 29	0.14	0.14
RSS before correction	4.08 (out of spec)	1.59

Procedure:

1. Bond the ring on an arbor, before the wave washer and cap are fitted. Slide the carrier (608 seated on its lip) onto a ground Ø8 arbor in a lathe or V-blocks. Spread epoxy on the register. Rotate and tap the ring until a dial indicator on the OD reads ≤ 0.02 TIR. Set the ring top 0.60 ± 0.03 mm above the 608 outer-race top face with a flat that has a 0.60 mm annular spigot (r 9.6–10.9) resting on the race. The collar top (25.5) is 0.1 mm below the flat, so it needs no relief. Cure undisturbed.
2. Static balance on knife edges or a magnetic balancer (use a second, freshly degreased 608 or spin the arbor).
3. Correct on the bottom face at r 31: Ø2 × 1 dimple −0.93 g·mm; empty M2 tapped hole −1.49; brass grub +1.35; W-alloy grub +2.70. Iterate until there is no preferred rest angle in 20 random releases.
4. Fit the wave washer, bond the cap ring, and fit the ESD strap.
5. Dynamic check (assembled): spin to 15 rev/s, log the LIS2DW12 at 200 Hz. The 1×/rev component must be ≤ the value measured with a 2 g·mm test mass taped at r 31.

Detent and encoder gap tolerance stack. The track height relative to the board is set by the hub length (turned ±0.02), the 608 width/face offset including axial play (±0.05), the jig (±0.03) and the board's local flatness (±0.05). The magnet height adds the carriage/cup (printed, ±0.1).

Gap	Nominal	RSS	Worst case	Effect
Detent	1.00	±0.13	±0.25	±13 % detent torque
TMAG	3.13	±0.11		window 2.2–4.0

Measure and shim: after assembly, check the magnet-to-track gap with feeler gauges at the click position. Correct whole-rotor height errors with 0.05 mm brass shim washers between the hub top and the board (not Kapton: the hub face is the GND contact). Correct magnet-only errors at the magnet cup.

---

6. Magnet track

Item	Spec
Magnets	48 × NdFeB N52 Ø2.0 × 2.0 mm, axial, NiCuNi, all from one batch (± 0.05 mm, ± 5 % Br). Buy 60.
Layout	Alternating N/S, R_t = 29.0, pitch 7.5° (3.80 mm chord): 24 electrical periods = 24 detents. Zero DC field, so stray steel causes 10–100× less cogging than a same-polarity track.
Pockets	Ø2.10 +0.02/0, 2.05 +0.02/0 deep. Measure your batch and set the pocket to max Ø + 0.03–0.05.
Flushness	± 0.03 to the ring top face, set against a flat while the epoxy cures
Adhesive	low-viscosity 2-part epoxy (3M DP420 / Loctite EA 9466). Seal the cured track with a thin wiped film. Keep it ≤ 60 °C (N52; N48H for hot use).
Field at the sensor	Bt/Bz ≈ 15/18 mT peak at 3 mm, zero mean ([detent research, magpylib]). The TMAG5273A2 ±133 mT range is ample.
Centrifugal load	5 mN per magnet at 10 rev/s, 49 mN at 30 rev/s, taken by the pocket wall
US sourcing	K&J D11-N52 (1/16″ × 1/16″ = Ø1.59 × 1.59) has ≈ 50 % of the volume of a Ø2 × 2 disc, so expect roughly 40–50 % less detent torque and encoder field. Do not substitute: order true Ø2 × 2 mm discs from a vendor that ships to the US (e.g. supermagnete, first4magnets, AliExpress).

Polarity jig. Neighbours 3.8 mm apart attract hard, and a flipped magnet gives a missing detent.

1. Print a 2 mm mask ring (PA12) that covers every other pocket (24 open) and clamp it on the ring top face.
2. Put a pin-head of epoxy in each open pocket. Stack the N52 supply on a steel rod north up and slide them off one at a time with a brass/plastic tweezer into the pockets.
3. Press a PTFE-taped glass plate (200 g) on top so they cure flush.
4. Swap to the second mask and repeat south up.
5. Check with magnetic viewing film (48 alternating spots). After assembly, the firmware calibration spin checks for 24 uniform electrical periods.

Keep-outs around the track (SPEC §7, detent and sensing research):

• No ferromagnetic part within 4 mm of the magnet faces, and none within 10 mm of the track or the TMAG. The stepper breaks this rule (A12).
• No power inductor within 10 mm.
• No pours on B.Cu or In2 at r ≥ 26.
• All screws near the rotor are brass or nylon; 316/A4 cold-headed screws are often slightly magnetic.

---

7. Detent mechanism

Principle. A stationary Ø3×2 N52 magnet sits 1.0 mm above the alternating track. Its strength is set by moving it radially inward, in plane. Both end positions are magnetically force-neutral, so the actuator holds at zero power.

Offset (mm)	0	0.5	1.0	1.5	2.0	3.0	4.0	5.0	6.0	7.5
Peak torque (mN·m)	16.5	13.7	11.4	9.5	5.7	2.0	0.53	0.14	0.029	0.0027
Rim force at r 35 (N)	0.47	0.39	0.33	0.27	0.16	0.06	0.015	0.004	–	–
In-well f_n (Hz)	9.2	8.4	7.6	7.0	5.4	3.2	1.6	0.8	0.4	0.1
Full steps from click	0	33	67	100	133	200	267	333	400	500

The bold columns are the verified LUT points; the rest are log-linear interpolations (calcs.py §4). The firmware default is full click (s = 100 %, x = 0, about 16.5 mN·m by the LUT; firmware.md §5.1), where the friction shoe damps it. An offset of 1.5 mm gives 9.5 mN·m. An EC11 encoder is about 10–12 mN·m; the EC11-like s = 85 % (x ≈ 1.15 mm, 77 full steps) gives about 11 mN·m, inside that range, and becomes the default only if the factory-measured shoe_x_max is at least 1.3 mm (firmware.md §5.1). The drawn shoe window (1.5 mm, below) meets that, so 85 % is the expected default on a unit built to it. In FREE (7.5 mm) the magnet's residual cogging is 2.7 µN·m. That is not the whole free-mode cogging: the stepper's steel can and lead screw sit 3.1–4.6 mm from the track (A12), and their cogging is unknown and possibly comparable to the bearing drag. Do not expect "no preferred rest angle" until §15 #2 passes.

Carriage (stl/detent_carriage.stl + stl/link_plate.stl).

• Body: printed PA12 (MJF) or POM, with a magnet cup for the Ø3×2 N52 (face 1.0 mm above the track) and a 6 × 5 × 0.48 plate that slides between two rails bonded to the board bottom.
• Capture: the plate runs in a 0.50 mm gap between the board and the rail lips, PTFE tape on both faces. Shim the rails with Kapton so the vertical play is ≤ 0.02 mm (feeler). Side clearance to the rail uprights is 0.05 mm. There is no spring preload: a preload below the ±0.7 N alternating normal force would not stop rattle anyway, and one above it costs ≥ 0.14 N of friction.
• Link: a 0.8 mm FR4/G10 plate (cut from PCB stock, non-conductive) bonded under the carriage web, with the 1.6 mm inclined slot (slope 0.75). In-plane deflection at 0.4 N is 0.03 mm, against 0.56 mm for the old 0.5 mm PA12 strip.
• Zero-lash pin: the slider's Ø1.5 brass pin is pressed against one slot flank by a small phosphor-bronze leaf on the slider, at 0.5 N. That is above the ±0.36 N slot-normal component of the ±0.29 N reversing radial force, so the slot tolerance does not matter.
• Screw backlash: use the VIC-TECH slider with the spring-preloaded (anti-backlash) nut, and check total lash at the carriage ≤ 0.03 mm with a dial indicator and ±0.5 N reversal (§15 #6). A separate ≥ 0.3 N anti-lash spring toward PARK (the verifier's option) does not fit as a near-constant force in this space. A real extension spring (0.35–0.65 N over the stroke) drops the thrust margin below 1× (table below), so it is the fallback only with a stronger actuator.

Actuator: VIC-TECH VSM08102. Ø8 PM can-stack linear stepper with slider, M2×0.4 lead screw, 10 mm stroke, 20 Ω/phase, ≥ 80 gf (0.78 N) at 500 PPS, voltage not stated (VIC-TECH). Driver: 2× DRV8837 (SPEC §5.6).

• Placement: chord at y = 15.5, axis z 23.8. The can is at x 6.5–15.5, z 19.8–27.8, cradled in a printed saddle (stl/stepper_saddle.stl) screwed to the θ 60 board hole. The lead screw points toward −x.
• Drive: the slider pin drives the inclined slot: 10 mm of slider travel gives 7.5 mm of carriage travel (1.33× force advantage). That is 0.015 mm of carriage per full step and 500 full steps end to end.
• Self-locking: the M2×0.4 lead (4.2°) is below its friction angle (≈ 9.8°).
• Why the 10 mm sibling is not an option: its Ø10 can does not fit the 10.1 mm bowl (web top 19.0 to board 29.1).

Force budget (calcs.py §10). The terms are: magnetic radial 0.29 N (opposes retraction); rail friction on N_v 0.70 + shoe 0.20 (+ tangential 0.55 while turning); the cam's lateral reaction F_x = 0.75 F_y on the rail sides; the shoe pad's radial drag; the shoe-lift ramp (8°, flexure 0.2 N/mm, 1.5–2.3 mm zone); and the 0.5 N pin preload.

Case (μ rail / pin / shoe)	Carriage F_y	Slider thrust	Margin on 0.78 N
Retract at rest (0.10 / 0.10 / 0.2)	0.57 N	0.57 N	1.36×
Retract at rest (0.15 / 0.20 / 0.3)	0.70 N	0.88 N	0.88×
Engage at rest (high μ)	0.22 N	0.35 N	2.3×
Retract while the rotor turns (high μ)	0.79 N	0.99 N	0.79×
Engage + 0.35–0.65 N anti-lash spring (high μ)	0.95 N	1.16 N	0.67×

So the 8 mm motor is marginal (≈ 0.9–1.4× at rest) on a rating given only at 500 PPS. The motor-drive simulation is stricter: it finds that the verified worst-case actuator load of 0.8–1.15 N exceeds what the VSM08102 delivers at any battery VSYS (0.93 N best case), so architecture risk R4 stays open: cut the preloads to ≤ 0.5 N or move to a stronger actuator (the fallbacks below; the 10 mm sibling does not fit) (hardware/sim, MD-4).

• Keep friction low: PTFE tape on the rails, lubricated slot, ramp ≤ 8°.
• Firmware: move the carriage only with the rotor at rest or below 1 rev/s, and run the loaded zone (0–4 mm) at ≤ 446 PPS (firmware.md §5.3; the SPICE motor tests check the 0.5 N budget at 446 PPS, hardware/sim, MD-3). Non-park moves MUST start only from VSYS_est ≥ 3.54 V (VSYS_est = 0.979 × the idle VSYS reading, so a reading ≥ 3.6159 V; firmware.md §5.8, §12.4), because at the old 3.40 V the stroke-load sag leaves the coils below the motor's 3.3 V / 165 mA rated drive (hardware/sim, MD-2).
• Gate: slider thrust ≥ 0.85 N at 3.0 V / 300 PPS against the real carriage (§15 #6). The bench test measures at both 300 and 500 PPS, which bracket the 446 PPS firmware speed. If it fails, fall back to the N20 or a Ø6 planetary micro gearmotor (below).

Stroke time (calcs.py §10): 267 steps in the loaded zone at the §15 #6 bench speeds (500 / 300 PPS) plus 233 steps at 800 PPS = 0.83–1.18 s (1.6 s at 200 PPS). The firmware profile (446 PPS in the loaded zone, then the 521–781 PPS ramp) takes about 0.9–1.1 s (firmware.md §5.3).

Packaging:

• The actuator sweeps θ ≈ 35–160°. The PCB rule is A11: bottom parts in the corridor must be ≤ 1.3 mm tall.
• Measured on the current PCB (calcs_output.txt §11): J40 (motor connector, 2.95 tall) clears the slider nut by 0.35 mm; every other bottom part, U40/U41 included, clears the actuator by ≥ 1.0 mm.
• The 33 mm envelope comes from a listing, not a drawing. Request the vendor drawing before the PCB is ordered.
• Ferrous parts: the can comes within 3.1 mm of the nearest track magnet and the lead screw within 4.6 mm. This breaks SPEC §7 (10 mm) and the 4 mm research keep-out (A12).

Fallback actuator: N20 3 V gearmotor + M3 lead screw (example). It has tens of N, uses the same connector (pins 1–2), one DRV8837 and DRV5032 end sensing. Its 10 × 12 mm section needs a board cut-out to sink 1.6 mm into the main board. A Ø6 planetary micro gearmotor with an M2 lead screw fits better (to verify). Both have steel too (A12).

Home sensing. DRV5032FA on the board top at (r 21.5, θ 90°), its die ≈ 2.4–2.8 mm above the parked magnet: more than 50 mT at park, −1.0 to −1.4 mT at click (firmware.md §5.4). The sensor is omnipolar with BOP min 1.5 mT, so on some units HOME may assert near full click (architecture risk R35). Home by a gentle stall at the force-free park end: the stall is the position reference, and HOME only confirms it on every return to FREE. Factory test 19 checks that HOME is released at full click.

Damping shoe (ringing). A high-inertia rotor in a conservative magnetic well is badly under-damped.

Release at 6° in a well	Bearing friction only	+ 1.6 mN·m shoe	Rest error with shoe (T_f/k)
Full click, the firmware default (16.5 mN·m)	156 cycles, > 20 s	4 cycles, 0.50 s	±0.23°
1.5 mm offset (9.5 mN·m)	121 cycles, > 20 s	2 cycles, 0.42 s	±0.40°

With the shoe engaged, a 5 rev/s click spin lasts ≈ 2.2 s (prize-wheel feel).

• Pad: a 3 × 0.8 mm wool felt or UHMW-PE pad (not PTFE, whose μ is too low) on a flexure at the carriage's outer tip. At click it lands on the ring top face at r ≈ 31.6–32.4, outside the magnets.
• Force: 0.20 N normal → 1.3–1.9 mN·m at μ 0.2–0.3, r 32.
• Engagement window: offsets ≤ 1.5 mm, which covers the full-click default (x = 0). An 8° ramp on the rail lifts the flexure 0.5 mm between 1.5 and 2.3 mm, so the pad never wipes the magnets. Detail the ramp with the real flexure.
• Firmware also predicts the landing detent.
• Felt wear, debris and μ drift: to verify on prototype.

---

8. Press-click mechanism

Kinematics.

• Rest: three springs (θ 60/180/300) push the FA plate up against three M2 brass pan-head stop screws that thread into the tops of the guide pins. Preload is 4.0 N total; the FA weighs 1.81 N, so 2.19 N net holds it on the stops: no rattle and no dead load on the switch.
• Click: the silicone press pad in the spindle's solid bottom (face z 12.75) closes the per-unit pre-gap, presses the TS-1187A, and the switch clicks at 0.45 mm of FA travel.
• Overload: at 0.50 mm the plate's three under-bosses land on the cup columns/ear (stop faces at z 11.5).
• Location: three Ø3 brass pins in Ø3.05 reamed holes, ±0.025 mm (0.03 mm at r 35, the detent-lash limit).

Part	Spec
Switch	XKB TS-1187A-B-A-B (LCSC C318884), 5.1 × 5.1 × 1.5, 1.6 N, 100 k cycles. Travel 0.25 ± 0.1 (to verify: measure every unit, reject > 0.32 mm). The C&K KSC2 family is 3.5 mm tall and does not fit.
Springs ×3	302 SS, wire 0.32, Dm 3.63 (OD 3.95), 3 active + 2 closed/ground, L₀ 4.72, k ≈ 0.63 N/mm. Installed 2.60 (ear/column seat z 10.5 → plate pocket z 13.1) → 1.33 N each. At the 0.50 stop: 2.10 mm, 1.65 N, τ ≈ 524 MPa (Wahl). Solid 1.60 mm → 0.50 mm margin (0.35 worst with solid +0.05 and seats −0.10).
Press pad	Ø3.0 × 1.0 mm silicone 40 Shore A, bonded with silicone adhesive in a Ø3.6 × 0.45 pocket in the spindle bottom (it protrudes 0.55 below the Ø7.8 relief, free to bulge). k ≈ 20–25 N/mm (to verify on a scale + dial indicator). The Ø7.8 × 0.5 relief clears the switch body corners.
Pre-gap	per unit: g0 = 0.38 − t_u (t_u = measured switch travel; e.g. 0.13 for 0.25), set with 0.025/0.05 mm Kapton discs under the pad. Measure it as the centre free play with the FA up against the stops.
Switch load at the stop	F_sw + k_pad × 0.05 = 2.7 N (k 22), 2.35–3.1 N for k 15–30; ≤ 3.9 N with a +0.025 shim error. Pad strain ≈ 12 %.
Guide pins ×3	Ø3.0 knurled brass, pressed + Loctite 638, plus a Ø1.0 brass cross pin pushed radially from outside the cup (θ 60 at z 8.0, θ 180/300 at z 6.0; fill the hole end with epoxy). Pins are 11.0 long (θ 180/300) / 7.0 long (θ 60), with M2 tapped 3 deep at the top. At θ 60: PEEK pin, PEEK cross pin, nylon screw (A4).
Stop screws ×3	M2 × 3 brass pan head (head Ø3.8 × 1.3), no washer, head top z 14.8 (0.8 mm under the ring at rest, 0.3 mm pressed). The head clamps on the pin top, which is flush with the counterbore floor (z 13.5); the head's outer annulus is the stop face.

Pull-out. After a flat drop the FA rebounds off the stop faces and is caught by the stop screws, which puts the pins in tension. Rough estimate: ≈ 230–800 N per pin for 0.1 m (restitution 0.3–0.6, stop stiffness 5–15 kN/mm, very uncertain). Capacity (knurl + bond at ~3 MPa, plus the Ø1 cross pin ≈ 160 N, PA12 bearing-limited):

Pin	Capacity
θ 180/300 (8 mm engaged)	≈ 390 N
θ 60 ear (4 mm)	≈ 270 N; less with PEEK

The old 1.2 mm bond (20–55 N) is gone. Pull-test the first printed cup (§15 #12).

Click force vs press point (calcs.py §7). The rigid FA is solved with unilateral stops (θ 180 stop at r 30).

r (mm)	over a stop (θ 60)	between stops (θ 0)	over the r 30 stop (θ 180)
0 (window centre)	4.64 N	4.64	4.64
20	2.65	3.00	2.76
29 (diffuser)	2.10	2.54	2.19
33.5 (ring top rim)	1.91	2.36	1.98

The force falls smoothly from centre to rim, about like a Surface Dial, with ±10 % variation round the rim (three-point stops).

Accidental clicks. A rim click needs only ≈ 2 N, so a hard flick can also click.

• Firmware ignores BTN_N while the rotor's angular acceleration is above a threshold (a press while the dial accelerates is a flick).
• If user tests still fail, raise the preload to 4.6 N total (rim ≈ 2.2–2.6 N) with longer springs (L₀ 5.03, same wire and coils) at the same installed length. The solid margin stays at 0.50 mm. Do not shorten the installed length: the old advice (2.3 mm) would have coil-bound the springs before the click.

---

9. Antenna placement and RF options

Placement (SPEC).

• Antenna: Molex 146153-0100, 34.9 × 9 × 0.1, dual-band, ground-plane independent, ≈ 3 dBi (Molex), on the printed shelf at z 5.5, centred at (0, 23.5), long axis along x (SPEC). Stick it with its own adhesive; its cable end is toward −x (verify on the Molex drawing).
• Distances: 9.9 mm below the ring bottom; 1.75 mm in plan from the pack side (SPEC, A1); 0.75 mm under the θ 60 guide ear (A4).
• Coax (1.13 mm, 100 mm): from the feed it rises at (−14.6, 19.9), outside the base board (r 23) and inside the FA ring (r 26). It crosses over J3 at z ≈ 15, enters the spindle through the θ 120 window/slot together with the harness, goes up the bore and out of the θ 270 slot and hub C-gap. It then threads between U12 and the module's inner corner (0.84 mm to U12, 0.44 mm above the rotating cap ring at r 9–13; tack it up with Kapton tape) and runs back beside the module at z ≈ 26 to the outward-facing u.FL (r ≈ 23). Modelled path 79 mm (calcs.py §12). Stow the remaining ~21 mm as one gentle loop against the cup wall in θ 100–150°, away from the antenna body (to verify S11 with the final routing).
• Plug passage: the MHF/u.FL plug body (≈ Ø2.0 × 1.3 plus the crimp sleeve, verify) must pass the 5.0 × 1.9 bottom window and the 5.0 × 2.45 top slot. Trial-fit the real plug in a spare spindle before machining the batch. The plug waits inside the bore while the rotor goes on (§12 step 9), so it never has to pass the 608.
• Bends: the turns into and out of the Ø6 bore are tighter than the cable's usual 5 × OD minimum (to verify; a 0.81 mm coax variant would help). Leave ≥ 3 mm of slack where the coax crosses the 0.5 mm press gap.

Metal near the antenna (check with a NanoVNA on the real build):

• θ 60 guide set, 0.75–9 mm above the antenna's +x end: use PEEK pin, PEEK cross pin and nylon screw from build 1. The 302 spring is a 68 mm wire coil, close to λ/2 at 2.44 GHz. If S11 shifts with/without the FA fitted, try a polymer spring of the same rate (to verify availability) or move the guide (SPEC change).
• Base board (4 layers) covers the antenna's inner 4 mm, 3.75 mm above it. SPEC keeps planes and pours on B.Cu, In1 and In2 out of y > 15.4 (B.Cu/In2 tracks and vias may come to y ≈ 17.5) and F.Cu pours out of y > 17.
• LiPo pouch (aluminium laminate), 1.75 mm in plan (A1).
• Stepper can (steel), 14–22 mm above the antenna's inner edge.
• USB-C shell is at θ 0, far away.
• The continuous 316L ring (circumference ≈ 1.6 λ), 10 mm above and partly over the antenna, is the main detuning risk.

RF options if S11 or range is bad, least to most invasive:

1. Retune with a π-match on the RF lead, if one is added.
2. Swap the θ 60 hardware (above) and move the pack away: a ≤ 30 mm-wide pack gives 4.0 mm.
3. RF-transparent ring: PA12-GF or PEEK ring of the same envelope with 16–24 discrete non-magnetic W-Ni-Cu slugs or ZrO₂/WC-Ni balls (Ni binder, not Co), carrying the same magnet shim. I ≈ 0.5–1.0·10⁻⁴ kg·m², so the spin time scales the same way — but note the ceiling. Matching the steel ring's inertia in this envelope means matching its effective density of 7.48 g/cm³, which needs ≈ 35 % of the ring volume filled with tungsten (lead would need 62 %, brass 86 %, steel 95 %). The top of that 0.5–1.0 range is the practical limit, not a design target.
4. A segmented 316L ring: 4–8 sectors with ≥ 0.5 mm insulating gaps (re-balance per build).

Validation:

• S11 in the closed device, hand on and off, with and without the θ 60 set.
• BLE RSSI at 1/3/10 m against the target (+8 dBm, 15 dB fade margin).
• Spin-down with the coax routed (it must touch nothing that rotates).

NFC tag coil (13.56 MHz, SPEC §1/§5.2/§8). SPEC fixes three things: the coil's place (in the top cover), its inductance (~1.4–1.5 µH, which C16/C17 180 pF tune) and its pads. The pads are J11 (NFC_COIL, no part fitted), flat pads on the main-board top at r 21.5, θ 222°: 1.2 × 2.0 mm at 2.0 mm pitch, at x −15.3 / −16.7, y −15.1 / −13.6. They lie under the OLED glass and the bracket rim. Everything else below is chosen in the model and tagged TVP there (to verify on prototype).

Section on the J11 axis (θ 222°, the tail's centre line). r grows to the right. The glass edge is oblique to this plane: the closest tail-to-glass gap (0.34 mm) is at the tail's edge. On this plane the sheet is cut away between r 16.8 and the fold line at r 24.7, where the tail tongue came from.

• Why not beside the glass: between the glass's lower corners (r 25.03) and the cover's six locating keys (r 26.4) there is only about 1.4 mm, and a recess there would cut the keys off the cover's centre part. So the coil sits above the glass instead.
• Coil: an annular 2-layer polyimide flex, ID 31 / OD 52.5 × 0.15 mm, black coverlay, with its copper turns in the band r 24.9–26.15 (4 turns of 0.20/0.15, or 3 turns of 0.25/0.25).
  • It is bonded with 0.05 mm PSA to the cover's ceiling at z 35.0, so it sits at z 34.80–34.95: 0.1 mm above the OLED glass and 1.05 mm under the cover's top face.
  • The cover's underside is raised to z 35.0 over the whole disc r ≤ 26.35. This replaces the rectangular glass pocket, which lies inside it. The cover therefore no longer boxes the glass in, and the glass is placed on the rim with a corner jig.
  • A 0.6 mm wall ring (r 26.35–26.95) keeps the six keys attached to the cover's centre part. The sheet's inner edge is hidden under the rim of the Ø30 window.
• Tail: a 2.0 mm wide tongue cut from the sheet at θ 222° (two traces 0.4/0.4), widening to 3.4 mm over the pads.
  • It folds down at r 24.7 (inner fold radius 0.45 mm; the tongue carries one copper layer) and drops vertically at r 24.10–24.25. That is 0.34 mm outside the glass edge (x −16.9) and 0.54 mm outside the bracket rim.
  • At the board it folds inward and runs flat (z 30.75–30.90) under the rim onto J11, where its lands are lap-soldered on the pads.
  • The path from the joint to the fold line is 5.6 mm. The developed tongue is 7.9 mm long, so its slot in the flat sheet runs from r 16.8 to r 24.7.
• J11 relief: the bracket rim's underside is raised from z 31.8 to 32.3 over J11 and the tail (±2.0 mm across the pads, r 20.1–24.7 along θ 222°), which leaves 0.5 mm of rim under the glass. The J11 joint body (pads + solder + tail end, 0.35 mm tall, from extract_board_parts.py) then has 1.25 mm of headroom instead of 0.75 mm, and the tail's board run 1.40 mm.
• Metal around the coil (calcs.py §13):
  • The In1 GND plane covers the whole coil, 4.35 mm below the copper (top-layer copper 4.10 mm). It is the main detuning and shielding element.
  • The OLED glass is 0.10 mm below the copper. The copper band overlaps the glass footprint only at its two lower corners, by 0.13 mm. A metal back on the panel, if it has one, would be 1.8 mm below the copper at those corners and 6.4–9.4 mm away in plan elsewhere.
  • The nearest LED lead frame is 3.5 mm away, the 316L ring and the track magnets 9.3 mm (behind the board) and the brass M4 nut 21.6 mm. The plastic diffuser and light box are 0.9 and 1.0 mm away.
• Inductance and tuning (estimate): the modified Wheeler formula gives 2.37 µH in free space for 4 turns (1.33 µH for 3). The image in the GND plane takes N² × 39 nH, which leaves ≈ 1.75 µH over the board (0.98 µH for 3 turns).
  • 4 turns would want C16/C17 ≈ 154 pF; the fitted 180 pF tunes 1.50 µH. Measure L in the closed device and retune (§15 #21).
  • If the read range is short, a 0.1 mm ferrite sheet under the copper band fits the 0.1 mm gap to the glass. It raises L, so C16/C17 drop to ≈ 95–110 pF (display-study.md §3).
• Press stroke: the coil, its tail, J11, the bracket, the glass and the cover all move together on the FA, so nothing in this group moves relative to anything else. The clash check finds the coil and the tail clear of the board, its top parts, the bracket, the glass, the light box and the cover.
• Assembly: solder the tail before the bracket goes on, and lay the sheet in last (§12 step 14). Range and detuning remain architecture risk R39.

---

10. Drop robustness

The 608's pure-axial static limit is 0.25·C₀ = 342 N (verifier). Rigid-seat rotor load for a flat drop onto a hard surface:

Drop	0.5 mm stopping distance	1 mm	2 mm
0.10 m	263 N	131 N	66 N
0.30 m	788 N	394 N	197 N
0.75 m	1971 N	985 N	493 N

Measures:

• Compliant outer-race seat. The wave washer (30–50 N, ≤ 300 N/mm) holds the race down from above, under the bonded 316 cap ring. In an upright drop the carrier moves down 0.15 mm and lands on the brass drop-stop flange (stationary, clamped in the spindle stack). The balls see at most 52–95 N, plus the race's own inertia.
• Inverted drops (landing on the cover) load the rigid lower lip. Option: a second wave washer under the race (double-compliant seat; to verify).
• Feet: the micro-suction pads add ~0.3–0.5 mm of stopping distance.
• Replaceable 608: remove the cover, optics, OLED bracket and M4 nut; lift off the board and hub; lift off the rotor. Warm the cap ring (~120 °C) and lift it; the 608 slides out of the H7 bore.

Other parts:

• NdFeB magnets are brittle; they are flush, epoxied and sealed.
• OLED glass: the Ø30 window is open over the glass, so a press on the window goes into the cover glass. The 0.7 mm chemically strengthened cover glass (simply supported on Ø30) deflects 0.025 mm at 4.4 N and 0.11 mm at 20 N, against a 0.6 mm air gap. Window presses therefore go through the cover glass and the cover skirt to the board, not into the OLED. (The old optional 0.5 mm PC lens deflected 1.65 mm at 4.4 N and pressed on the OLED.)
• The FA is stopped by the cup columns/ear in a drop, and by the stop screws on the rebound and when lifted by the cover (pull-out, §8).
• The battery sits on foam tape, with no sharp features nearby.

Recommended tests: 10 drops from 10 cm onto wood, flat and on the edge, then check spin-down (≤ 10 % loss), the click and the pin pull-out.

---

11. Feet and anti-slip

A 10 rev/s flick puts 0.21 N·s of lateral impulse into the base, about 7 N over 30 ms. Plain rubber gives 1.3–2.2 N of static friction, so the puck slides unless it is held or stuck.

Option	Behaviour
3 × 70° micro-suction arc pads, r 25–31 (model)	Stick to glass, laminate and painted wood. Tip edge 28.1 mm: tip force 11.6 N at the rim. Recommended.
3 point feet at r 28 (SPEC)	A rim press tips the device at 1.60 N, below the click force. Do not use.
High-μ gel feet	The device tips instead of sliding under a flick (verifier). Avoid.

---

12. Assembly order

1. Ring. Measure OD/ID TIR and face runout. Do the N52 cling test (§5 process order). Clean the pockets with IPA.
2. Magnets. Mask-jig procedure (§6); check with viewing film.
3. Bearing. Degrease and oil (§4); spin-test and pick the best of 5.
4. Rotor.
   1. Bore/turn the printed carrier (Ø22 H7 + register, one chucking).
   2. Slide the 608 into the collar against the lower lip (slip fit).
   3. Bond the ring on the Ø8 arbor: OD TIR ≤ 0.02, top 0.60 above the 608 top face (flat with the 0.60 spigot). Cure.
   4. Statically balance (§5).
   5. Fit the wave washer, bond the 316 cap ring on the collar top, bond the PB ESD strap with conductive epoxy (ring ID → web → collar → under the cap). Check ring-to-outer-race continuity with a DMM.
5. Cup.
   1. Heat-set the M2 × 3 brass insert into the θ 270 boss.
   2. Press the three knurled guide pins with Loctite 638 (perpendicular within 0.1 mm at the top, with a printed square), then push in the Ø1 cross pins from outside and seal the holes with epoxy. θ 60 uses PEEK (A4).
6. Antenna. Stick the 146153-0100 on the shelf at (0, 23.5), cable end toward −x. Lead the coax to the riser point (−14.6, 19.9) and tape it.
7. Battery. The pack is re-terminated with 28 AWG pre-crimped SH leads, two per rail, in an SHR-04V-S housing (pins 1–2 BAT+, 3–4 GND; SPEC §6). Foam tape on the floor; pack against the θ 0 end stop. Lay the leads from the θ 180 end beside the pack at z ≈ 10 (they pass under the spider) to the board edge at θ ≈ 222°. Check polarity before plugging in (SPEC §6).
8. Base board. Lower it onto the θ 90 rib with the tab 1.5 mm short of the wall, slide it out into the USB slot, then fit the brass M2 screw at θ 270 (r 20). Bring the battery leads up over the board edge and plug them into J2 (r 17, θ 222°, mouth radially outward).
9. FA lower stack (on the bench).
   1. Press the silicone pad into the spindle bottom with the per-unit Kapton shims (§8; measure the switch travel on the base board first).
   2. Push the spindle up through the plate's D-recess, then slide on the brass lower spacer (D-bore, window at θ 120).
   3. Thread the harness and the coax into the bore now, but keep their top ends inside it. The harness is 12 pre-crimped single wires: SSH-003T-P0.2 contacts, 5 × 28 AWG for pins 1–4 and 8 (SPEC §2) and 7 × 32 AWG. The SHR-12V-S housings (~14 × 3 mm) cannot pass the slots. Push the contact ends in through the θ 120 window/slot and ~10 mm up the bore with a pull string, then the coax, plug first. Cut the wires ≈ 60–70 mm (modelled path 43 mm with J1 at r 19.5, plus slack for plugging J3 before the plate goes in).
   4. Insert the base-end contacts into one SHR-12V-S housing.
10. Suspension. Plug the housing into J3. Drop the springs into the column/ear pockets and lower the plate over the pins (wires and coax through the open sector at θ 120). Fit the stop screws now, while nothing covers the counterbores (no washers, brass-compatible threadlocker). Check the centre free play (= the per-unit pre-gap), the click at the centre (4–5 N) and the hard stop at 0.50 mm.
11. Rotor onto the spindle. Lower the rotor so the 608 inner race slides down to the brass spacer. The wires and coax inside the bore do not obstruct it. Then fish the wire ends and the coax plug out of the θ 270 top slot (above the inner race).
12. Hub and main board (pre-assembled).
    1. On the board: bond the rails (shim to ≤ 0.02 mm capture), fit the carriage + FR4 link + magnet, fit the stepper and saddle (M2 brass at the θ 60 hole) and plug in the motor.
    2. Slide the brass hub down the spindle with its C-gap over the wires (D-flat at θ 90).
    3. Lower the board onto the hub over the M4 spigot.
    4. Insert the main-end contacts into the second housing and plug J1. Plug the u.FL (module end facing outward, r ≈ 23). Tack the coax up with Kapton tape where it passes between U12 and the module's inner corner, so it stays clear of the rotating cap ring below it (0.44 mm modelled); spin the rotor by hand to check.
    5. Fit the M4×0.5 brass nut, Loctite 222, 0.10 N·m with a pin spanner.
    6. Check the magnet-to-track gap (1.0 ± 0.1) and shim the hub with 0.05 mm brass if needed (§5).
13. First power-up. Flash over the TC2030 (pads under the glass, so before step 14), home the actuator, run the encoder calibration spin and the detent f_n self-calibration, the spin-down gate (≥ 35 s; target ≥ 45 s) and the FREE-mode rest-angle histogram.
14. Optics. Bond the light box (septa between LEDs). Glue the diffuser into the cover (clear UV adhesive) and bond the 0.7 mm cover glass into its recess.
    1. NFC tail. While J11 is still uncovered, fold the coil sheet back outward over the light box, lay the tail's lands on the pre-tinned J11 pads and reflow them with a fine iron (§9).
    2. Bracket. Fit the OLED bracket with 3× M2 × 6 brass CSK screws through the board into the hub. Its rim relief goes over the joint, and the tail leaves under the rim's outer edge.
    3. Glass. Fold the FPC into the ZIF and stick the glass on the bracket rim (0.2 mm foam). Centre it with a corner jig, because the cover no longer boxes it in.
    4. Cover. Fold the coil sheet back over the glass and peel its PSA liner. Lower the cover and snap it onto the light box, then press once over the glass border so that the PSA takes (the glass foam gives the 0.1 mm).
15. Feet. Apply the arc pads.
16. Final tests. Click force at the centre and 3 rim azimuths; BLE RSSI at 3 m; NFC tag read with a phone; ringing settles < 1 s at the default strength; ESD path < 10 MΩ; drop test on a test unit.

---

13. Purchasable mechanical BOM

Prices are rough 2026 single-unit estimates. "NM" means it must be non-magnetic.

The five machined parts (items 4, 7, 8, 9, 10) are ordered from hardware/mechanical/step/*.step and hardware/mechanical/drawings/*.pdf, not from stl/: no CNC vendor accepts an STL, and a solid cannot carry a fit, a thread, a knurl or a general tolerance. drawgen.py builds both from the same constants as the model — see hardware/mechanical/README.md.

#	Item	Spec	Qty	Source / part	Est.	Notes
1	Main bearing	SKF 608-2Z (alt. NSK 608ZZ, SKF 608-2RSL)	1 (+4)	bearing distributors; SKF 608-2Z	$2–8	degrease and oil (§4)
2	Bearing oil	ISO VG 2–10 synthetic instrument oil (Mobil Velocite No. 3 / No. 6)	1	industrial suppliers	$10	base oil to verify
3	Wave washer	single-turn, 22 mm bore, ID ≥ 19.5, working height ≤ 0.5 mm, 30–50 N, ≤ 300 N/mm, stainless	2	Smalley / Rotor Clip / generic; check the catalogue height	$1–4	many catalogue parts work at 0.8–1.5 mm: they do not fit. Load to verify.
4	Cap ring	0.5 mm 316 sheet, OD 24.4 / ID 19.9 (laser-cut or a shim washer; drawings/cap_ring.pdf, step/cap_ring.step)	2	laser-cut service / shim-washer assortment	$1	NM
5	Track magnets	NdFeB N52 Ø2 × 2 axial, NiCuNi	60	supermagnete, first4magnets, AliExpress (one batch)	$3–6	not K&J D11 (§6)
6	Detent magnet	NdFeB N52 Ø3 × 2 axial	3	as above; near-equivalent K&J D21B-N52 (1/8″ × 1/16″, slightly weaker)	$1
7	Flywheel ring	316L, 70/54/10, turning only: chamfers 1.0 top-outer / 1.5 bottom-outer / 0.5 bottom-inner, Ø62.00 × 2.00 magnet-shim recess in the top face, 4× M2 balance holes at r 31, straight knurl z 17.1–24.1, process order §5 (drawings/rotor_ring.pdf, step/rotor_ring.step; the knurl is a drawing callout, not modelled: with it the model gives 109.5 g, §14)	1	JLCCNC / PCBWay CNC / local	$52.49 quoted (JLCCNC, SUS304, qty 1, 2026-10-20: machining $41.05, material $6.91, engineering $4.53, no finish). Was $71.45 with the 48 blind pockets, so moving them to the shim (item 7a) saved $18.96, machining −32 %. 316L not offered there; get a PCBWay quote	magnetism check: JLC warn 304 may become weakly magnetic after cold working
7a	Magnet shim	Printed Ø62 / Ø54 × 1.95 ring, 48 Ø2.10 through holes at R 29.000 every 7.5°; drops into item 7's recess and is located by it (stl/magnet_shim.stl, §5)	1	JLC3DP / PCBWay MJF PA12 or SLS PA12-GF	$0.32 quoted (JLC3DP, qty 1, 2026-10-20, 9600 resin; print it in item 35's batch)	0.95 mm walls are under JLC3DP's 1.2 mm recommended wall, though its 0.8 hard minimum is met: prototype first. Magnet angular position is the open risk, §5
8	Spindle	turned 316L from Ø12 bar: Ø8 h6, bore Ø6 (z 14.45–28.25), flange Ø10 × 0.6 with D-flat, D-flats at θ 90 (z 12.8–18, 25–29.05), slots 5.0 wide at θ 120 (z 14.45–16.35) and θ 270 (z 25.8–28.25), pad pocket Ø3.6 × 0.45 in a Ø7.8 × 0.5 relief, M4×0.5 spigot Ø4 to z 32.3 (drawings/spindle.pdf, step/spindle.step)	1 (+1)	CNC turning service	$15–30	NM; check with an N52
9	Hub	turned C360 brass, Ø17 flange × 3.0 + Ø11.5 × 1.1 spacer, D-bore, 3× M2 tapped 2.6 deep at r 6.5 (θ 90/210/330), 5.0 mm C-gap at θ 270 (drawings/spindle_hub.pdf, step/spindle_hub.step)	1	CNC	$8–15	no heat-set inserts (a 0.4 mm wall would split). The model's bore is still round — the D is in this spec and on the drawing as an OPEN note, not in dial_spinner.scad; cut it in before ordering
10	Lower spacer	turned C360 brass, Ø11.5 × 3.7 + flange Ø21.2 × 0.45, D-bore, 5.0 window at θ 120 (drawings/lower_spacer.pdf, step/lower_spacer.step)	1	CNC	$5–10	replaces the bonded POM spacer. The model's bore is still round — same open D-bore as item 9
11	Nut	M4×0.5 brass, OD 7, 1.5 thick, 2 pin-spanner holes (or a thin DIN 439 M4×0.5 turned down)	2	CNC / fine-thread nut suppliers	$1–3	0.10 N·m, Loctite 222
12	Press pad	Ø3.0 × 1.0 silicone 40 Shore A	5	punched from sheet	–	k to verify
13	Guide pins	knurled brass Ø3.0, 11.0 mm (×2) and 7.0 mm (×1), M2 tapped 3 deep at the top	3	brass rod + knurl, or Ø3 brass dowels	$1	NM; PEEK Ø3 at θ 60 (RF)
14	Cross pins	Ø1.0 brass (×2) + Ø1.0 PEEK (×1), 7 mm	3	rod	$1
15	Press springs	302 SS, wire 0.32, OD 3.95, 3 active + 2 closed/ground, L₀ 4.72, ≈ 0.63 N/mm; spare set L₀ 5.03 for 4.6 N preload	3 (+6)	Lee Spring / Associated Spring near-match, or custom	$0.1–1	measure the rate
16	Stop screws	M2 × 3 brass pan head (DIN 84); 1× nylon for θ 60	3		$0.5	NM
17	Bracket screws	M2 × 6 brass countersunk (DIN 963) into the brass hub	3		$0.5	NM
18	Saddle / base screws	M2 × 4 brass pan head	2		$0.3	NM
19	Heat-set insert	brass M2 × 3 (OD 3.2)	1		$0.2	base θ 270 boss
20	Antenna	Molex 146153-0100 (100 mm 1.13 coax, MHF/u.FL)	1	DigiKey / Mouser; -0050 = LCSC C916292 is too short	$2–3	§9
21	Battery	protected 1S LiPo, 503450 / 523450 class (cell 5.0–5.2 × 34 × 50, ~900–1000 mAh), ≤ 5.6 mm max incl. PCM and tolerance, ≤ 34.5 wide, ≤ 55 long incl. PCM (≤ 53 preferred, A3), leads from the PCM end at ≥ 4 mm off the centreline, re-terminated for J2 (item 36). Alternative for RF: a ≤ 30 mm-wide pack (A1).	1	generic protected 503450 with a vendor drawing (EEMB/PKCELL class, verify)	$4–8	The old EEMB LP603449 (6.3 mm) breaks the ≤ 6.0 limit. Verify polarity.
22	OLED	Newhaven NHD-1.5-128128G (alt. Orient AOC128128A0-1.50W)	1 (+1)	Newhaven, LCSC C17322695	$15–28	long lead time
23	Cover glass	0.7 mm chemically strengthened glass, Ø33.0, anti-scratch	1 (+1)	round smartwatch cover glass (AliExpress) or waterjet	$2–5	bond with UV/epoxy in the recess
24	Detent actuator	VIC-TECH VSM08102 Ø8 linear stepper + slider (anti-backlash nut), M2×0.4, 10 mm	2	VIC-TECH	$3–7	ask for the drawing, thrust curve and a JST SH lead
25	Fallback actuator	N20 3 V gearmotor + M3 lead screw, or Ø6 planetary + M2	1	example	$3	§7
26	Link plate	0.8 mm FR4/G10 (stl/link_plate.stl outline; cut from a PCB panel or order as a 0.8 mm PCB)	2	PCB fab	–	non-conductive
27	Pin preload leaf	0.1 mm phosphor bronze, 1.2 × 3	2	shim stock	–	NM
28	Press switch	XKB TS-1187A-B-A-B (on the base PCB)	–	LCSC C318884	–	measure travel
29	Damping shoe	1 mm wool felt or UHMW-PE tape (e.g. 3M 5423) on the printed flexure	1		$1	not PTFE
30	Feet	micro-suction tape 1.0 mm, 3 × 70° arcs, r 25–31	1 set		$2
31	Diffuser	opal PMMA 1.8–2.0 mm, laser-cut ring r 27–30	1	laser-cut	$2
32	Link harness	12 single pre-crimped SH wires (SSH-003T-P0.2 both ends, no housings): 5 × 28 AWG (pins 1–4, 8), 7 × 32 AWG, ~100 mm; 2 × SHR-12V-S housings	1	JST distributors / pre-crimped SH lead sets	$4	insert into housings after routing (§12)
33	ESD strap	0.1 × 3 mm phosphor bronze, ~25 mm; silver conductive epoxy (e.g. MG 8331)	1		$10
34	Adhesives	2-part epoxy (3M DP420 / Loctite EA 9466); Loctite 638, 641, 222; UV glue; silicone adhesive	–		$25
35	Printed parts	stl/: base_cup, top_cover, light_box (black, interior painted white), fa_lower_plate, rotor_carrier (PA12-GF, machined after printing), magnet_shim (item 7a), detent_carriage, oled_bracket, stepper_saddle, carriage_rail	1 set	JLC3DP / PCBWay MJF PA12 or SLS PA12-GF	$25–45	FDM for fit checks only
36	Battery leads	4 single pre-crimped SH wires, 28 AWG (SSH-003T-P0.2), two per rail, + 1 × SHR-04V-S housing for J2 (SM04B-SRSS-TB: pins 1–2 BAT+, 3–4 GND)	1	JST distributors / pre-crimped SH lead sets	$1–2	replaces the pack's own leads/connector; check polarity
37	NFC coil	Custom 13.56 MHz flex: 2-layer PI, Ø31/Ø52.5 × 0.15, black coverlay, 4 turns 0.20/0.15 at r 24.9–26.15 (≈ 1.75 µH over the board, estimate), 2.0 mm tongue tail with 3.4 × 2.2 lands for J11, 0.05 PSA on the top face (§9)	1 (+1)	flex-PCB house, drawing from the model (no quote yet)	–	retune C16/C17 for the measured L (§9)

---

14. Calculation results (summary)

Full output: calcs_output.txt. Re-run with python3 hardware/mechanical/calcs.py.

Rotor.

Part	Mass (g)	I (kg·m²)
316L ring (net, recessed)	108.3	1.065e-4
48 magnets	2.26	1.9e-6
Printed magnet shim (PA12-GF)	1.37	1.2e-6
PA12-GF carrier (no ribs)	5.55	2.1e-6
608 rotating share	7.65	8.1e-7
Wave washer, cap ring, strap, adhesive	1.4	4.3e-7
Total	126.6	1.129e-4

The shim (§5) costs 5.3 % of the rotor's inertia against the solid-pocketed ring's 1.192e-4; §4 has what that does to spin-down.

Energy and momentum.

rev/s	E (mJ)	L (mN·m·s)	Rim speed (m/s)
5	56	3.55	1.1
10	223	7.10	2.2
15	502	10.64	3.3

Spin-down: §4. Detent and ringing: §7. Balance: §5.

Press click: §8. Centre 4.64 N, rim 1.91–2.36 N; springs 0.50 mm from solid at the stop; switch 2.7 N at the stop.

TMAG5273 sampling vs the 24-pole-pair track at 30 rev/s (TMAG5273 datasheet).

Mode	f_s	Electrical advance / sample	Nyquist speed	Margin over 30 rev/s
XZX, 4× avg (sensing research config)	3.1 kSPS	83.6°	64.6 rev/s	2.15× (1.94× at −10 % oscillator)
ZX, 4× avg	4.4 kSPS	58.9°	91.7 rev/s	3.06× (2.75×)

XZX averages two X conversions about 215 µs apart, so at 30 rev/s X shrinks to 0.88 of Z: up to 3.5° electrical (0.15° mechanical) atan2 error. Correct it in firmware as a function of speed, or use ZX at high speed (estimate).

Mass budget.

Group	Mass
Base	43 g (cup 12.0, LiPo 19, base board 6.9, guide hardware 2.3, rest 2.5)
FA without rotor	50 g (main board 12.3, optics + cover glass 8.5, OLED + bracket 5.0, brass hub + nut 4.7, spindle 3.6, stepper group 3.9, brass spacer 2.6, plate 2.4, 608 inner share 5.3, rest 2.0)
Rotor	134 g
Total	227 g, CoG at z 19.4. Static tip angle on the arc pads ≈ 55°.

Detent actuator: §7 table. Margin 0.88–1.36× at rest, stroke 0.83–1.18 s.

Wiring and windows (calcs.py §12).

Item	Value
Fill, 12 wires + coax	bottom window 51 %, top slot 40 %, hub gap 32 %, bore 17 %
Harness path (current PCB)	43 mm
Coax path (antenna feed to the outward-facing u.FL)	79 mm (21 mm of the 100 mm cable left to stow)
Battery leads path (pack end to J2)	21 mm
Cover glass deflection, 0.7 mm	0.025 mm at 4.4 N, 0.11 mm at 20 N
USB-C overmold 6.5 / 7.0 mm tall: radial clearance to the ring	1.12 / 0.87 mm at rest; 0.62 / 0.37 mm pressed

Cable constraint: USB-C plugs with an overmold taller than 7 mm can touch the spinning ring while the dial is pressed.

NFC coil and tail (calcs.py §13, estimates).

Item	Value
Coil	copper r 24.9–26.15 at z 34.80–34.95, 4.35 mm above the In1 GND plane
L, 4 turns (3 turns)	2.37 (1.33) µH free; ≈ 1.75 (0.98) µH over the GND plane
C16/C17 for 4 turns	≈ 154 pF (the fitted 180 pF tunes 1.50 µH)
Tail to the glass edge / bracket rim / rim relief roof	0.34 / 0.54 / 1.40 mm

---

15. Prototype validation checklist

#	Item	How to measure	Pass
1	Bearing friction / spin-down	Encoder log of a 10 rev/s flick	gate ≥ 35 s; target ≥ 45 s; record the value
2	Free-mode cogging from the stepper can + lead screw (A12)	Before the PCB order: ring on its 608 + the actual VSM08102 held at its modelled position (can 3.1 mm from the track), 50 slow-release rest angles, with and without the motor	no 48/rev preferred angle; else change actuator or layout
3	Detent LUT at R_t 29 / top face	In-well f_n at offsets 0 / 1.5 / 3 mm → T1 = I(2πf_n)²/24	within 10 % of §7
4	Click-mode rocking (single 608, gravity preload)	TMAG Bz ripple; listen	inaudible; else preload option (A5)
5	Shoe engagement window, felt μ, wear	Ringing time at default strength; 10⁴ engagements	< 1 s, no debris
6	VSM08102 envelope, thrust, lash	Vendor drawing; push-scale on the real carriage at 3.0 and 3.3 V, 300 and 500 PPS; dial indicator with ±0.5 N reversal	thrust ≥ 0.85 N at 3.0 V / 300 PPS; lash ≤ 0.03 mm
7	TS-1187A travel (every unit), pad rate, click force vs radius/azimuth	Dial indicator; scale; force gauge at 0/20/29/33.5 mm, 3 azimuths	travel ≤ 0.32; k_pad 15–30 N/mm; §8 ± 15 %
8	Spring rate and preload	Scale + calipers	1.2–1.5 N each at 2.60 mm
9	Antenna S11 and BLE range (1.75 mm to the pack, θ 60 hardware, ring)	NanoVNA, RSSI at 1/3/10 m	S11 < −10 dB at 2.40–2.48 GHz
10	Coax and u.FL plug pass the slots; bend; press slack; Kapton tack between U12 and the module	Trial fit in a spare spindle; 10⁴ press cycles; feeler/visual check under the board	passes, no kink, no drag; coax ≥ 0.3 mm above the cap ring
11	Balance	LIS2DW12 1×/rev at 15 rev/s	≤ 2 g·mm
12	Guide-pin pull-out	Pull the pin (via the stop screw) on the first printed cup	≥ 300 N (θ 180/300), ≥ 200 N (θ 60)
13	Drop	10 × 10 cm onto wood	spin loss ≤ 10 %, pins hold
14	Ring magnetism after knurling	N52 cling test	none
15	Light-box uniformity (2.6 mm air gap)	Photo at mid brightness	acceptable segments
16	Battery envelope incl. PCM and swelling	Calipers, after 50 cycles	≤ 55 × 34.5 × 6.0
17	Detent / TMAG gaps after assembly	Feeler gauges	1.0 ± 0.1 / 3.1 ± 0.3
18	ESD bleed path	Megohmmeter ring → GND, rotor turning slowly	< 10 MΩ
19	Wave washer load and height	Scale + calipers	30–50 N at 0.5 mm
20	XZX speed-dependent ellipse	Encoder error vs speed on a turntable	< 0.2° after correction
21	NFC coil fit, tail route to J11, tuning, read range (R39)	Trial fit with the bracket, glass and cover on (tail modelled 0.34 mm from the glass edge, joint 1.25 mm under the rim relief); LCR meter on the coil in the closed device (estimate ≈ 1.75 µH for 4 turns); Android and iOS phones	tail clear of the bracket rim and the glass, sheet flat on the cover ceiling; C16/C17 retuned to the measured L; tag reads on both phones

---

16. Model files and how to regenerate

File	Purpose
hardware/mechanical/dial_spinner.scad	Parametric model. Views: assembly, section, section_xz, exploded, board_bottom, base_top, clash, clash_geom, nfc (section on the J11 axis), prims, part
hardware/mechanical/extract_board_parts.py	KiCad boards → board_parts.scad / board_parts.json (needs KiCad 9 pcbnew)
hardware/mechanical/calcs.py	All calculations in this document; writes renders/calc_*.png. §11 and §13 (NFC coil and tail) read the model via view="prims".
hardware/mechanical/check_clashes.sh [margin] → clash_check.py	CGAL interference check (rotor, press, rest, FA-internal pairs); legend in clash_legend.txt
hardware/mechanical/export_stl.sh	Exports the 16 printed/machined parts to stl/
hardware/mechanical/parts_plate.scad	Overview render of the STLs (renders/parts.png)
hardware/mechanical/render_all.sh	Regenerates everything, including renders/nfc.png (about 7 min)

cd hardware/mechanical
python3 extract_board_parts.py              # after any PCB change
xvfb-run -a openscad -o renders/assembly.png --imgsize=1200,900 \
  --camera=118,-136,104,0,0,17 --colorscheme=Tomorrow -D 'view="assembly"' dial_spinner.scad
openscad -o stl/top_cover.stl --export-format binstl -D 'view="part"' -D 'part="top_cover"' dial_spinner.scad
xvfb-run -a openscad -o /tmp/free.png -D 'view="section"' -D 'det_offset=7.5' dial_spinner.scad   # free-spin state