The problem for me is how broad the rate system is. It's like almost 20 options, with a 0-10 on each. It's exhausting trying to determine that for each girl. I think it needs to be a bit simplified. Or, it needs to be sectioned. Like lvl 1 is only judging their face. Then lvl 2 is judging their body. lvl 3 is judging their personality. lvl 4 is judging their private parts.
Something like that.
Yes this. Also there are 800+ actual parameters you are trying to optimize, and you can see all of them in the save editor. 10 bars to control 800+ parameters is difficult. Possible, but you have to understand the problem and work the model, tuning how many parameters you will allow, how the input (rating bars) are attached to parameters, etc.
I'd have to disagree here. While the rating system is complex, it is really the only reason this game has such diverse girls and so much replayability compared to other games.
It's not really an opinion thing on what is wrong with the system.
Genetic algorithms are good for having low inputs (just the dozen or so ratings) and high parameter counts (the 800+ features you see in the save editor), but you still need to "iterate" a sufficient amount of times if you want to have high accuracy. The concept is fine in a broad sense, but the author isn't tuning it. "Iterate" here means rating. The iterations/ratings take an enormous amount of time with the "interview" system for the player to rate the dozens and dozens of girls you need to get what you want. Given we have 800+ parameters and only 10 ratings, the genetic algorithms work better when you have a computer do it for you as fast as possible, possibly thousands of times, which would be like you playing this game for 500 hours.
This isn't the best application of a genetic algorithm or really any machine learning. I think the author just wanted to toy with the concept and misapplied it here. I think it is salvagable, but real core things have to be dealt with.
However, unlocking new traits as you level up sounds like a good idea.
Unlocking traits would be the same as interviewing a lot of girls but only rating using a few bars, leaving the rest zero, if I understand what the game is really doing (it counts a zero as "leave the attached parameters as randomly generated"). Rate only head, face, and hair, leave everything else zero. Do that 30+ times. Then only rate breasts, hip/waist, legs. Do that 30 times. Etc. You can do this now, but its obvious this is still too slow and you still need many, many iterations to make it work. If rating could be done in <5 seconds, removing all the fumbling around with the computer and overlays, and used a very simple 2D interface that was significantly faster it might work. Right now, you have to move around, click "wait a while" or "next day", wait for the fade out, load screen, fade it, move your character around, fumble with the rating overlay again. It's too slow, the cycle time is far too slow to make that viable right now.
The author is doing the wrong things. Autorating and cheats are basically working around the problems instead of dealing with them head on. As great as the wide variety of girls that can show up is, the real problem is the parameter count is far too high for how many iterations you can get.
1. The parameter count has to drop. Combing shoulder, arm, forearm, hand, and combing thigh, foot, calve, etc. These are OBVIOUS fixes to reduce parameter count and have no gameplay value. I feel it would be fairly trivial to cut the parameter count by 75% and not really affect the variety in the game,
unless you just feel having girls cum instantly when you touch their forearm but get instantly 100% pissed off if you graze their shoulder is compelling gameplay. I don't think it is compelling gameplay at all, and the constant confusion by players seems to be evidence it is not good gameplay.
2. The iteration speed has to increase. The player needs to be able to rate more girls much faster. Autorating was an attempt to do this, but I don't think it really speeds it up much because its more about time fumbling around in the 3D space or having to slowly and painfully and carefully touch the 3 pixels of the "knee" to see if she'll either want to kill you or cum instantly, and is still the wrong approach. Maybe just rendering a dozen girls, you click portaits to fill your "queue" of girls, something like that would work better, or improve the UI/UX so you can quick press buttons to go to the next girl, make dismiss instant/faster, remove a bunch of forced delays in the conversation system, etc.
3. It's really a combination of both of the above. The fewer parameters you have, the fewer iterations needed to "focus in" on the result the player wants, and vice versa. Again, I feel parameter count is the bigger problem, and there are so many super obvious fixes here.
Autorating system was simply a mistake on time and effort by the author. The cheats were also just a bad workaround for the 98% chance of glass or angry girls that should've been a proper normalization or remapping of the parameters involving anger/sensitivity behavior. The core gameplay can be fixed with attention to the correct fixes. Right now, every new player is only frustrated by the fact they can rate for dozens of hours and still get bad results, and have to have constant detailed instructions on how to rate to make it maybe work. Or players end up having to use save editors, or load other people's saved girl cards
to completely skip the core gameplay. 
