You absolutely cannot get 17 questions wrong in Verbal and get a V47 score. A V47 scorer will normally have two mistakes in Verbal. Even if you answered all your experimental questions incorrectly, there aren't enough experimental questions to get you to 17 mistakes. But experimental questions cover all difficulty levels, so most of them will be easy for a V47 test taker - a V47 test taker will almost never answer more than one of them incorrectly, and will most often answer all of the experimental questions correctly.
Note that the Quant section is different. You can have a lot of mistakes and still get very high Quant scores. The sections are different because the scaled scores translate differently to percentiles - a Q50 score is not the same thing as a V50 score. It is really the same, in percentile terms, as a V39 score. And just as the Q50 test taker can afford to make several mistakes on hard questions, so can the V39 test taker. The Verbal scoring scale really goes a lot higher than the Quant does, since the Verbal scale is capable of differentiating among test takers deep into the 99th percentile (any score between V45 and V51 is in the 99th percentile, while a Q51, the highest possible Quant score, is only in the 97th percentile). If the Quant scale instead went up to 60, so the test distinguished between test takers above the 97th percentile, then to get the Q58-Q60 scores, you would have almost no margin for error, just as you do now in Verbal if you're aiming for a V49-V51. But the Quant scale stops at 51, and to get a Q50, you can make a lot of mistakes, as long as those mistakes are mostly on the hardest questions.
Anyway, there is definitely a bug in the scoring of that test, and that graph they present you at the end is just not how the scoring algorithm works. Since they're not implementing the algorithm correctly, it's not surprising their scoring is off. If you want to get a good estimate of your Verbal level, there's no substitute for the official GMATPrep tests.