I took a look at the pre-LS2 and post-LS2 QA plots you sent me and the
Legacy MC vs Luminance MC vs pre-LS2 QA comparisons. I can summarize as
follows.

(1) The post-LS2 electronics lets the chamber response function into raw
data without processing, which is known to be power-law (long tail) due to
MWPC signal formation physics. The pre-LS2 electronics had an analog
signal shaper in order to get rid of this. The post-LS2 Shine has a
digital implementation for this same correction (the so-called Wiener
filtering). The post-LS2 QA plots are visibly obtained _without_ the
Wiener filter, therefore the number of time slices distribution looks
wide. If the Wiener filtere were applied, the time response is pretty
similar to pre-LS2 electronics (I had detailed study on this). See also:
https://twiki.cern.ch/twiki/pub/NA61/Nov2022Wiener/wiener.pdf
   and
https://twiki.cern.ch/twiki/bin/view/NA61/Nov2022Wiener
Thus, there are two possible strategies:
(A) Minimal working solution: do not update TPC time response in MC, but
then do not use the Wiener filter for MC reco, but use it for the post-LS2
raw data reco.
(B) Purist solution: write a TPC time response distorter for post-LS2 MC,
and use Wiener filter in all reco.
I think, we can live with (A).

(2) The pad response functions look (and should look) pretty much
unchanged after upgrade, in the raw data. If we are satisfied with what we
had pre-LS2, then no upgrade needed here.

(3) For the high multiplicity data, for the sectors having lots of tracks,
Piotr usually lowers the gains (SenseWire HV). This affects VTPC1, VTPC2,
and the high resolution sectors of MTPCs. I am not sure if this is
disturbing for MC. In my opinion, if we don't have manpower, we can
postpone to simulate this, as it has little effect on cluster/track
finding, but more on dE/dx only.

(4) Geometry: this has changed. I think one of the big issues with MC that
it is difficult to track where it takes various detector descriptions
from. Ideally, it should take everything from det::Detector container. If
I remember well, for the geometry this is taken from some custom root
file. I am not familiar with the procedure here, I just remember that it
was a mess and not well documented (Marek Szuba et al).

I think, what is important is to simulate those effects in which we can
count on large calibration imperfections, regarding cluster/track finding.
Of course, we have to fix the geometry. Then, the most important one is
e.g. ExB distortion. Then, the E distortion. The cluster shape simulation
is the least important one, I think, if we correct the raw data with the
above machinery. Of course, it would be nice to simulate everything, but
the above would be my priority list if we are lacking manpower.

Best,
Andras