Take me with a grain of salt, because I haven't worked with Java in a while, haven't made any bots, and all my stats knowledge is from one class from a couple years ago.
If you roll a dice, there is a equal chance to get any number, but if you roll two dice, you have a higher chance of getting a seven than a two. Most things work more like the second one, where not all outcomes have a uniform chance of happening. Something with a non-uniform distribution and no direct cut-offs is more humanlike. With probability distributions, there are two important graphs. One is called a PDF, or Probability Distribution Function, and the area under it shows the probability for an interval. The other, is called the CDF, or Cumulative Distribution Function, and it shows the probability up to a point. I was doing this method, I would ignore the harp percentage and only base it on time. I'd play around to find a good normal distribution,
for example, use a mean of two minutes and a standard deviation of ,7 , then use that. Then pick a random float from 0-1 and plug that into the inverse of the CDF to get the wait time until we click again. This method may not have as good xp because it wouldn't look at the harp percentages, (it would click on average of x minutes, so sometimes it tunes early, and sometimes later, and rarely would stop or time out) but it would probably be harder to detect because it is a more natural randomness.