While this is of course possible, think of the probabilities. What the OP is saying is true - it’s very unlikely what happened to him as of course while there is a 50% chance of it happening each time, the chance of it happening twice would be 50%*50% =25% and do this all the way up to 23 times in a row and you are getting into the tiny probability of 1 in 4 million odd as he mentioned.

Try flipping a coin 23 times and see how long it takes you to get 23 heads in a row, it will take a very long time! These odds 50% are the same based on what the OP said. Same as doing the same leveling he did but getting a special level every single time…

Again, you are right that with a small sample size, you can get anomalies and to create a meaningful analysis you need much more data however op is definitely right to say what he got is way out of expectation assuming all done correctly. If it happens again then something is definitely wrong because the odds on that would be astronomical.

Pseudo RNG will use seed numbers and a formula to create a distribution of outcomes that are in line with what you would see if you did a truly random event, except that you could recreate the outcome of you know the seed number. Therefore although this is only pseudo RNG the outcome as described should still be programmed via the formula used to happen once every 4m times. So you could be very unlucky, you could have misinterpreted something when setting up your experiment or, if you can recreate a potential bias with loads and loads more data, you could start to say the RNG could be wrong, or at least wrong on the assumption it was designed to mimic pure randomness…