Where the Magnet Goes
The constraint
The AS5600 resolves angle from the direction of a field it assumes is uniform across four Hall elements on the die. It needs a diametrically magnetized disc centered over the package. There is no off-axis mode; an off-axis ring does not degrade gracefully, it goes non-monotonic.
The tilt rod passes straight through the shell, a Ø11 through-bore in both halves. So there is no shaft end to put a magnet on.
The crown gear solves that. A 12-tooth module-1 spur fused beside the worm wheel drives a 16-tooth crown gear, which turns the rod's horizontal rotation onto a vertical axis and creates the shaft end the rod refuses to give you — with the magnets on that axis, over the die. The crown sits under the rod, beside the worm, teeth up at the spur. It is a sensor pickoff, not a drive element. It is also the most important printed part in the set, because its tooth quality is the angle noise floor.
Easier to drag than to describe:
Tilt rod
20T wheel + 12T spur · the bar is the slat plane
0°
Crown gear
seen from above · magnet in the center pocket
0°
- Rod
- 0°
- Crown
- 0°
- Encoder
- 768 / 1536
- Motor
- 2.50 rev
- Per count
- 0.117° at the rod
- Wrap headroom
- 2.67×
TiltForge
Direct LAN control
Position 50% · open
encoder 768 / 1536 counts
State IDLE · Calibrated yes
Or drag either dial. A full sweep runs 5.0 s, the real tilt time.
A simulation of the page the firmware serves, driven by the gear ratio — not a live device.
Real ratios; the page beside it is the one the firmware serves. Standalone at /drivetrain.
The numbers
- 12T into 16T is a 0.75 step-down. Over a full 180° tilt the crown turns 0.375 of a revolution: 1,536 of the encoder's 4,096 counts.
- Resolution at the rod: 0.117° per count. A blind does not need that; backlash and clutch hysteresis will limit you long before the encoder.
- Headroom before the absolute reading wraps: 2.7×.
- Misalignment budget: AMS allows ±0.25 mm radial. The whole stack RSSes to 0.274 mm — about ±1.1° at the crown, ±1.5° at the rod. Invisible on a window blind. I would ship it, and I am.
The magnet is two magnets
The obvious part — the Ø6 × 2.5 diametric disc every vendor sells as "the AS5600 magnet" — is not what is in there.
Inside the crown's tooth ring, on its top face, is a Ø12.38 recess 2.00 mm deep. Two standard 3 × 6 × 1.8 mm blocks lie flat in it, side by side, poles opposed — 6 × 6 mm in plan, 1.8 tall. That runs the field across the bore the way a diametric disc does, out of parts any supplier stocks. An axial disc, same size, reads alive and means nothing.
They rest on a thin printed floor, and a Ø6.25 hole runs down through that floor to the sensor. So the magnets sit on top of the crown and the AS5600 looks up at them through the middle of it.
The gap the part already sets
The sensor is on the bottom copper of the Rev D board — and the board goes in inverted, so that face points up and the AS5600 sits directly under the crown.
The production part sets the air gap itself, which I did not expect. Measured off the STL: the magnets' underside sits 2.80 mm above the crown's bottom face, and that bottom face carries a Ø7.06 × 1.77 mm pocket. A SOIC-8 body is 4.90 × 3.90 × 1.75. The sensor nests up inside the crown.
So the gap is 2.80 − 1.75 = 1.05 mm plus the crown's running clearance over the board — a couple of tenths lands mid-window on the 1.0–2.5 mm the sensor wants, set by the printed part instead of by a shelf height someone has to hold to a tenth.

Rev D's bottom side, from the board render. U2 is the AS5600, C5 and C6 beside it. Installed, that face points up — and U2's body sits inside the crown's bottom pocket, with the magnets a millimeter above it.
And one trap: the AS5600 die is 0.295 mm off the board's geometric centerline. Register the shelf bore on the sensor's real position (10.455 mm from the long edge, 6.185 from the short edge), not on board center. A board-center assumption alone spends the whole ±0.25 mm budget.
Verify on the bench, not on paper: the firmware prints the AGC on boot, and mid-scale — about 64 on the 3V3 scale — beats every calculation above.
Two more items from the audit
The crown STL carried 761 non-manifold edges in August — a TinkerCAD boolean on a high-segment revolve, on the one part that sets encoder accuracy. Re-exported, the production part is down to 33. And the AS5600's one-shot burn commands are deliberately not implemented: the zero point lives in NVS, so it survives a sensor swap and can be recalibrated in the field.
The TiltRod, by TiltForge, is a retrofit smart blind motor — repairable, open, Matter-native, and the cord always works. Founders pricing on the first 100 units; first units ship November 2026. Check your headrail at tiltforge.com/fit or follow the build at tiltforge.com/#waitlist.
TiltForge — a retrofit smart blind motor. Repairable, open, Matter-native, and the cord always works.
Follow the build — new entries straight to your inbox