Steady mode

1 · Pick your build

The diagnostic builds do not sharpen — they measure. See step 4.

2 · What changed

Two things, both behind one switch:

• The gravity reference is smoothed before it steers the angle. A stroke pushes the blade sideways, and the old code could not tell that apart from gravity — 0.18 g of sweep tilted the reference 10.2° while changing its magnitude only 0.016 g, far too little to be rejected. Averaging the reference removes the stroke, which is oscillatory, and keeps gravity, which is not.
• The number and the colour hold against chatter. A reading hovering at 19.5 used to flip 19/20/19/20 forever, and a knife sitting on the tolerance edge strobed the whole screen and machine-gunned the buzzer. Both now need a real move before they change.

Simulated, on a 0.18 g stroke — measured on your stone is what actually counts:

3 · How to test it (the important bit)

You can switch it on and off on the device, so you are comparing against the same stick, same knife, same stone — not against your memory of last week.

1. On the TOLERANCE screen, hold B to flip STEADY: ON / STEADY: OFF. It sticks across restarts.
2. Sharpen a full side with it ON. Note how much the number moves and how often the colour flips.
3. Turn it OFF and sharpen the same way again.
4. Tell me: is the number calmer? Does the colour still change fast enough to be useful, or does it feel late?

Also new in this build: an accuracy check (hold A on SET TARGET — lay it flat, press A, then stand it on a known angle; there are printable wedges if you don't own an angle block), time on-angle on the summary screen, and your last five sessions (hold B on the summary). Session duration now survives Plus / Plus2 sleep-wake (it used to show 00:00). A crash or reflash no longer resurrects the previous knife as RESUME. Plus2 / S3 no longer treat the wake / power-on press as “off.”

The risk to watch for is lag. Smoothing always trades response for calm. If green now feels like it arrives after you have already found the angle, say so — the time constant is one number and easy to back off.

4 · Optional: get me numbers

The diagnostic build runs both filters — old and new — side by side on the same sensor data, so it can measure the difference on your hardware rather than in my simulation.

1. Flash the diagnostic build for your stick.
2. In the flasher dialog choose Logs & Console (or any serial monitor at 115200).
3. Hold the device still at your angle for the 2-second reference capture.
4. Sharpen normally for a minute and copy me a few lines.

Each line reads like this — sd and pp are how much the reading moved, so lower is better, and gain is how many times steadier the new filter was on your data:

raw sd= 1.42 pp= 6.10 | steady sd= 0.38 pp= 1.70 | gain 3.7x/ 3.6x | hdrm 0.0 smp 0.0mg | readfail 0/100

hdrm answers a question I could not settle without hardware: how many spare IMU samples sit unused between ticks, and how much a per-tick average would change the reading. If that number is large, there is a second, bigger fix worth doing. If it is near zero, there isn't.

Going back

The normal flasher puts the current release (v1.0.0) back on. Your tolerance and buzzer preferences survive the round trip as long as you don't tick the installer's erase option.