OK, trying to establish some good scalings. If I simulate some pupil data ranging from 0 - 8mm and plot, I get this:

sfreq = 100
ch_names = ["my_pupil"]
ch_types = "pupil"
info = mne.create_info(ch_names, sfreq, ch_types)
data = np.full((1, 900), np.repeat(np.linspace(0, .008, 9), sfreq))
raw = mne.io.RawArray(data, info, verbose=False)
raw.plot(remove_dc=False)

@sappelhoff do you have the bandwidth to try this branch out with your eyetracking data?

Thanks, this seems to work very well!

there are two pupil dilation channels on top, and two "eye openness" channels at the bottom. The default scaling on this branch gets it right automatically, whereas on main, I need to manually adjust using + and -.

This is the same data window on main:

Ok awesome!

Working off the last MWE, here are the epoch and evoked plots:

import numpy as np
import mne

sfreq = 100
ch_names = ["my_pupil"]
ch_types = "pupil"
info = mne.create_info(ch_names, sfreq, ch_types)

baseline = np.repeat(6, sfreq / 10) # 100 ms

# Create a parabolic pupil constriction response

x = np.linspace(0, 2, sfreq * 2) # 2 seconds
a = 4 
h = 1
k = 2
y = a * (x - h) ** 2 + k
ep_data = np.concatenate([baseline, y])
ep_data /= 1000 # convert to meters

# 10 epochs, 1 channel, 200 samples
epochs = np.full((10, 1, len(ep_data)), ep_data)
epochs = mne.EpochsArray(epochs, info, tmin=-0.1)
epochs.plot()
epochs.average().plot()

Which looks reasonable to me. I'm going to push an empty commit to trigger the CI's.

great, thanks scott!

@drammock Also just a heads up that the pip-pre jobs didn't pass the last time the CI's ran. The failures were in the decoding module

yeah that's happening in #12860 too so I'll go ahead and merge this one, we can chase down the decoding-on-Windows problem over there.