Bingo...
Disconnecting the probes, setting all the channels manually to 1:10, then adjusting the decoder threshold-levels.
Re-booting the scope, now the threshold values remain.
Siglent, we have a bug in the autosense circuit (programm?)...
And this seems to affect not only the level setting for the decoding, because I had already told early that when fall back from the bodeplot mode a channel is set back to 1:1, although a 1:10 autosense probe was connected and recognized as such before starting the bodeplot.
Or how the scope froze for me for the first time a few weeks ago, with the message that it had detected a probe on channel 3, although the probe was already connected.
So it seems to influence many things, a rather serious bug.