That is how multiple choice questions with 5 options work. Can confirm
If familiarizing yourself with the types of question or how to unravel the most common puzzle types for S3 works for you, great. But from memory, the evidence in favour of prep for S3 points to a relatively small potential gain at a population level (although undoubtedly some individuals find it more useful than others). Notably, said small S3 gain was counter-weighted by the *decrease* in S2 performance found in those doing paid prep courses - there were no significant differences in total UMAT score.
The research I'm thinking of was carried out a good 5 years ago now, but it was methodologically very sound and I doubt very much that either UMAT or the prep material has changed enough to invalidate it.
I disagree in general about there not being much evidence for S3 improvement. S3 largely mirrors a significant component of intellectual assessment in psychology, and there is a substantial body of research (and we all know, if you want research done right, you get a psych to do it

) to support practice effects in this to the point that IQ assessments re-administered within 12 months are generally frowned upon. Dedicating time to learning strategies and patterns is definitely beneficial.
The major methodological flaw with the small amount of (pretty poor) UMAT research I’ve seen has been a lack of pre and post prep intrasubject testing. Typically, the research has been looking at the Prep Population vs the Non Prep Population, and to some extent that’s irrelevant. I admit it would be hard to do (though definitely possible), but the best research would be to administer parallel forms of the UMAT to the same participant, first with no prep, and then following a period of prep, and comparing their results. You could also potentially determine the profiles of people (age, sex, years of education, other) that benefit, that don’t benefit, or that don’t need to prep, and compare the types of prep to see what is most useful/not useful at all.***
You can probably now start to see why ACER (an educational research body!) aren’t doing this research themselves — or, at least, aren’t making it public!
As I mentioned here earlier, I did an informal version of this on myself. I completed (the relatively ‘easy’) ACER 3 with zero UMAT knowledge in March 2016, and got 13/40-odd on S3, and < 50% of questions right overall. I had no idea how to manage timing or to approach most of the questions. Last year I got 98th%ile (inc. 64/100 on S3) on the more difficult ‘real’ UMAT. If I’d done the ‘real’ UMAT with no prep, I may as well have handed $250(?) to ACER and walked back out of the room!
*** if anyone is aware of this research please point it out to me, because I’m very interested in this topic.