Sample size varies greatly with what you are trying to learn and what sort of poker is played.
If the sample is for self reflection, you can get a reasonable meta-perspective in big bet games with 10,000 hands, perhaps less. But if you are playing limit poker, where the variance is much bigger than the win rates, you'll need at least a hundred thousands hands.
If the sample is for villain identification, you have to learn to live with the uncertainty of small or even tiny sample sizes. For example, say a villain sits at your full ring table and plays the first ten hands. What can you say about that? Could be he is posturing, though that is futile in a world with lifetime records. Could be he hit a super lucky run of starting cards. Could be he is quite loose. Thing is, you can relatively quickly make some overall assumptions about certain villains.
However, if you wonder about river bluffs vs value bets, you will be wondering a long time. How often do you get to see the villain's hand after he takes aggressive river action? It is relatively rare, you might need weeks worth of table time to get a couple of dozen data points. So here, you will be mostly using other poker skills as your HUD doesn't have even a marginal sample. You should lean in on your HUD reads for sure, but take into account the other information as well.
As for evaluating your play . . . . . better to post hand histories than HUD data. Please only one hand at time. And remember good results does not mean well played. Not does bad result mean poorly played. Pick hands that seem interesting to you or that caused you to ponder, remembering that a lot of money changes hands in pots where one pair was a contender.
Good luck -=- DrStrange