The 538 number is not a state polling average, it’s a highly modeled projection that involves demographics and economic numbers and national polls etc. To be fair, RCP probably isn’t a good baseline either, especially this year when they’re getting progressively more cherry-picky and ridiculous about their poll inclusion and exclusion choices.
Edit: That said, it looks like RCP excluded some polls that if they were included would have put the average at just about what the result was, so I guess you’re right. A great example of why not to trust RCP, I should have known better. I’m pretty sure Beto did actually over perform his polls though.
RCP seems to be repeating that mistake in Texas this year too, excluding most polls showing Biden leading here, like PPP and DFP