After one hand(or a small sample of hands), you can't really make a conclusion about the type a player is - in your example with 23o, they might just have a bad day and 'throw' away the chips when in most cases is actually a good overall poker player* this doesn't mean that you can't make a note about the 'weird' play but just don't make your future decisions against this player based on 1 previous experience... Still, if similar notes about the same player add up after a few sessions, than it's usually safe to say you have his type of poker player figured out and you can adjust your strategy based on it...
I really feel like theory is taking a back seat to practice in this type of response. I think if I didn't make future decisions based on conclusions made from the single hand like in my example, I would be leaving money on the table for the sake of a principle. Should it turn out that future information disconfirms my conclusions, then obviously I have to be able to change my assessment.
Put it this way. If my one-hand assessment is correct most of the time (and I think it is), and I act wisely accordingly (call shoves very wide in cash games, or when ICM considerations allow for in tourneys), then I'm going to make lots more money than I loose when my assessment is wrong. If I sit down at a table, and the same player raises my blind all in first hand, I'm calling with 10s, and I'm confident that I'm going to win that pot 3 out of 4 times. Obviously, if I see the player at the table fold 5 hands pre-flop first, or make normal sized opening bets for a while before making a preflop all in, then I'd be an idiot not to factor in that information. I just think it's dogmatic to not act on a pretty obviously displayed weakness just because it's
possible that your assessment is incorrect. It's always possible that your assessment is incorrect. Maybe you played a guy for months or years and he has displayed some exploitable weaknesses in his game, but had a great coach shore those holes up yesterday. The question isn't if you could be wrong, it's if it's likely.
It's no different a mindset from calling an all-in against a lag when you don't have the nuts (you're likely ahead, but maybe not). You're choosing to make an optimal decision, not a risk-free one. Do you really dispute that my example is going to produce an accurate assessment most of the time?