In both cases the noise is pulled up, not by Auto EQ but by the multiband.
Hans we checked this by loading 8.23 and it's settings (currently used) in Wavosaur Batch versus 9.01 with the same settings processed by Watchcat.
There is NO noise in Wavosaur results, but there IS bigly noise in the Watchcat results. It might very well be the compressor but it is only audible in 9.01 and was not in 8.23 using the same compressor settings (symmetrical setup in both cases), nor has this noise been audible in 7.84 's settings we used before 8.23 on live. This recording is one of our fixed tests.
Also I reported the latency issue at wavosaur's forum linking to this thread. I had some real problems understanding the issue here, but I now understand this is about every new record starting with the settings of the last processed vs default. But isn't this corrected in the blink of an eye by the ST software?
I heard a lot of noise too when I tested it here - it was there in the original, and - of course, given the settings - much worse in the processed version. Haven't tried 8.23 yet - should I just load the settings that you sent me earlier in 8.23?
Chances are that if you're testing in Wavosaur with 8.23 and 7.84, that the result completely depends on the song before it. If a new track starts, Auto EQ starts in the center - but if a previous song ended with not a lot of high, it will still be boosted a lot when the new track starts. And Auto EQ responds very slowly to changes - it can easily take 30 seconds or so before the effect is completely gone.
So about "in the blink of an eye": Depends. The multibands, maybe, especially if you're not using gating. The AGC is a very different story already. Auto EQ is extremely slow by design. That doesn't explain why you also hear the highs in WatchCat - but the other thing that definitely influences all this is the AGC. If the AGC was far down after the previous track, it can take long for it to get up. During that time, the multibands do less - so you also get less boost in the highs.