No Spring

A dispatch from the audit, August 28, 2026. One paragraph in the mesh audit explained six weeks of knife-edge clutch settings.

---

What holds the setting

The bench unit's clutch stack is flats and fibers, nothing else. There was no shaft length for the Belleville washers the design calls for, so the only thing holding the preload is the nylon insert in the locknut.

A nyloc retains position, not force. It stops the nut backing off. When the stack settles, the nut stays exactly where it is and the preload drops. The nyloc does its job perfectly and the clutch goes slack anyway.

Without a spring, the entire axial compliance of the joint is the elastic stretch of the M4 thread. At 26 N of preload over 10 mm of engagement that is 0.148 µm. Two Bellevilles in series give about 400 µm. That is a factor of 2,700.

Against 0.148 µm of available travel: fiber bed-in at 2% of 0.8 mm is 16 µm (108×), PETG creep under sustained load is about 25 µm (169×), and a 20 °C swing on 0.8 mm of PETG is 1.1 µm (7×). Any one of those takes the preload to zero. And it fails open: the clutch slips under normal drive, which a customer would report as "the motor stopped working." That is a bad field failure to diagnose, and it is exactly the failure this product exists to replace.

The problem was never that the spring was too stiff. There is no spring, which is infinitely stiff. With 0.15 µm of travel, a few degrees of nut rotation swings you from nothing to bound.

What has been running since August: fiber, gear, fiber, nut. No spring, because there was nowhere to put one.

Why the shaft is 18 mm

The revised Hotec shaft carries 18.0 mm of M4 thread. Here is what has to fit on it:

Item	mm
Inner thin nut (DIN 439)	2.20
Clutch plate A	1.60
Fiber washer	0.80
Pinion	?
Fiber washer	0.80
Clutch plate B	1.60
Flat M4×9×0.8	0.80
Belleville ×2 in series	1.50
Flat M4×9×0.8	0.80
Outer nyloc (DIN 985)	5.00
Subtotal	15.10

That leaves 2.90 mm for the pinion. If the pinion is the 4.5 mm from the earlier axle stack, the pack is 1.6 mm over, and the shaft is already made. So the pinion gets a caliper before the motors land, and if it is over, the stack gives: one Belleville instead of two saves 0.75 mm, a wave washer in place of both saves 1.0 mm. Either is still hundreds of times better than no spring. [VERIFY: measured pinion width.]

One rule survives whatever the stack ends up as: a flat washer either side of the spring, because the knurl must never bear on PETG.

Set it by turns, not torque

Nut torque is a poor spec: thread friction scatter means one wrench reading covers 21 to 35 N of actual preload. Once a spring is in, the target is about 0.021 N·m, below anything a wrench resolves.

The reproducible spec: run the outer nut to first contact with the stack, then advance a measured fraction of a turn. M4×0.7 means one full turn is 0.70 mm of travel, and two Bellevilles give about 0.4 mm, so the whole adjustment window is just over half a turn. That is unit-free, survives a spring change, and can be handed to anyone with a work instruction.

Then verify by measurement: stall the output and read the torque at which it slips, a known mass on a known arm at the rod. Not measured yet. [VERIFY: slip-torque measurement on a sprung stack.]

Until the sprung stack exists, the bench unit is a stopgap, not a durability sample. I am not drawing reliability conclusions from it.

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.