I'm using the latest versions of micropython stable edition, esp32-idf3-20191220-v1.12.bin, and it seems there is a problem with the neopixel implementation on the M5 Atom.

When I run the following program, there are random errors in the display of the pixels.

from machine import Pin, lightsleep
from neopixel import NeoPixel
np = NeoPixel(Pin(27),25)
black = (0,0,0)
while True:
    np.fill(black)
    lightsleep(10)
    np.write()

https://www.youtube.com/watch?v=vvI9tY6uXHI

From other places I hear this might be a problem with the power supply to the neopixel array, and might disappear if I supplied external voltage lower than 4.7V, because there is a problem with the voltage range on USB. That might work, it might not... but I wonder if there is a better solution, or a product modification, that might fix this?

http://micropython.org/download/esp32/
http://micropython.org/resources/firmware/esp32-idf3-20191220-v1.12.bin

@thrasher Yes that's what I was doing, and thought there must be a better way.
There are a couple of operations that require the 'download' method of installation, and will not work in the 'play' mode.

If I write any code in M5Flow 1.41-beta OR 1.32 and then "download" the code and run it on my M5StickC, there doesn't seem to be any way to get back to using M5Flow. The device appears to be completely disconnected from the network, not responding to further "download" commands, and can't even run code.

Solved
When the device reboots, instead of allowing it to reboot and run the downloaded code, hold down the reset button AND the button B at the same time, then the device will go back into a monitoring mode.
Is this documented anywhere?