Sequences of random numbers are streaky by nature. They have to be or they are not random. The current result cannot have any predictive value for the next result in a random sequence.
We’ve never been presented with any statistically significant evidence that the die roller has a problem. Every statistical test I’ve ever run on the the roller’s output has had a result consistent with generation from a true random source.
If we were presented with such evidence, that would very likely mean there’s a problem with the PRNG used by Java, because there’s only a line or two of our own code involved. That would be a huge deal because it’s supposed to be cryptographically secure and is relied on by many large enterprises. As it is, it’s undergone extensive scrutiny over the years (such as being FIPS-140-2 certified) and no problems have turned up. It’s not impossible that something is wrong despite all that, but it won’t be a thing anyone can just eyeball or detect from small samples.
For reference: