Weird behavior when manipulating phase-inverted layers

I recently upgraded to SpectraLayers Pro 9.0.0, System is Windows 10 Pro Update 21H2. I really like the upgrade so far, especially one thing I have been waiting for so long seems finally fixed.
I stumbled upon a weird behavior, though. When I soft-invert the phase of a layer via the Layers panel and do some manipulations, SL 9 processes the audio strange, if not to say “wrong”.

Here is a small visualization of what happens with the Eraser or Amplifier tool. For the purpose of better understanding what is happening, I just generated a simple 1 second long 10Hz mono square wave with -12dB peak. A 4000px square brush with 50% hardness and level adjustment of -12db/+9db was used to apply the tools (done on a 4k display).
As long as I only see the live preview of the brush, it seems all OK and logical, but as soon as I apply the Eraser or Amplifier, things get weird. When erasing, the level actually increases (although not by 12dB) and when amplifying the phase of the waveform gets slowly inverted, according to the hardness of the brush (until +6dB the level decreases; at +6dB the level seems to be fully gone; above +6dB the level increases again, but is phase-inverted). I couldn’t even guess why it is doing that…

I also tried the Gain process with negative and positive values as well, which seemed almost “right”, but just almost. For this I made a selection from 0.5 seconds towards the end, with a fade of 0.075 seconds.

The level gain is applied correctly, including the soft fade, but the whole thing is phase-inverted right from the start of the selection.

I checked my older SL Pro 8.0.1 again, and I would say that the behaviour here is the more “right” one…
Here for Amplify and Eraser tools…

… and here for the Gain process …

In my mind I just would expect SpectraLayers to do manipulations/calculations based on the actual sample/spectral data, regardless of the Invert Phase fuction in the Layers panel, just as it does in SL 8.0.1. Even it were supposed to use the phase-inverted data, then it doesn’t make sense to phase-shift during further processings again …
And just to make sure, as I never used SL 8.0.2, I did a quick test there and it showed the same issues as in SL 9.0.0. I also did not test any other tools or processes, to see if there is any strange behavior.

Could some developer please look further into this?

Thanks for reporting the issue, I’ll have a look at this !