Without intervening into the knock detection calibration, I logged some variables to see what's going on inside the knock detection routines. A sample attached below. I left only the most interesting things in the graph, the rest you have to trust me on, they would haze the picture or are beyond the magnified area.

It shows what happens more or less typically when I looked at the other parts of the logs. The green line that goes up to 1 at 123.5 s. indicates the rpm ramp up dynamic mode for knock detection is on (no wonder, the rpm gradient goes up to 4000 rpm/s). Dynamic mode = knock detection is dampened by a factor of ~1.2. 2 on the green line would mean load ramp up, but here it does not happen (though the load ramps up too). The violet line going up to 15 means that the frequency setting changes on all four cylinders going through the calibrated rpm thresholds and this also temporarily triggers dynamic mode. The rpm climb you see from 123.5 to 127.5 is a rolling quick pull in the 2nd gear. The knock retard on one cylinder (others were 0) kicks in when the load touches on the target one (EDIT: the retard kicks in already when the load starts to climb, initially it goes up at a rate of 180%/s, yet no dynamic mode). And this is on E70!

What bothers me is ngas, the rpm gradient. the blue thing. It says somewhere in the FRs that it is measured every 10ms (so defo more than one time per engine rotation at these rpms), nevertheless, I do not see why it's spiked and why a rather smooth rpm line gives such gradient readings. I can also see that the n/a setting I quoted above would turn the dynamic mode on much more in the particular area.

Seriously curios now if that's the way it's supposed to be...?

EDIT 2: I also looked at some other bins from the  family. The n/a version has also the load thresholds for dynamic mode more or less at the half level of my ECU = goes into dynamic mode more likely. One of the SS bins though has the rpm threshold slightly lower, while the load thresholds are seriously higher (around twice as much in the lower rpm range).

I think I have figured out my cold start / cranking issues for Ethanol, now I am only annoyed that I was one day too late with this - the proper frost we had here for the last two days is gone and we are back to 0*C temperatures.

It seems my biggest mistake / pitfall was that I considerably raised the target idle RPM at colder ranges, by around +150 at 0 degrees, 250 at -10, etc. Apart from getting the start up fuel and decay rate right it seemed all the time that the throttle chokes the process. Goes like - engine kicks in, gets slightly choky rpms at some level below 1000, some time elapses and it springs to life at the target rpm (which was ~1200). This hesitation drove me nuts, so I turned to FR for Alfa ME7.3 and it became less of a mystery. Without requoting the whole document - the choke period is the apparently the TVLLRSTE delay (2 seconds) from the end of start (reaching 800 rpm) until the idle PID kicks in. It can also kick in quicker, that happens right after the rpms reach the target idle rpm for the first time, but for me this never happened, hence this 2 second choke. And it never happened because I raised the targets. Now, I know that the maps should be clever enough to deal with this and the drag torque map should set the load at the right level, but clearly there must be something about Ethanol that makes it not so. So I tried (a) dropping the target idle speed slightly (but still higher than factory), and (b) raising DMDNSM - the after crank torque offset. The result is this:

https://youtu.be/Cligw9REpz4

So far a singular experiment, but it all makes sense and it seems the ECU has finally started to react to my wishes. Somewhere in the process I also played with the starting ignition both ways, but that seemed to do more bad than good.