Let us say you have a winrate of 5bb/100 over a 'large' sample

How many hands sample would it need to be to where if you lost 10 BI's in one day, the winrate would only change .5/100 to 4.5/100?

Because, I theorize, that is about which point a sample size is accurate enough to be considered a 'true' winrate.

The reason this question came to mind is that over a 7000 hand sample at 50NL, my winrate has fluctuated from -.09 to +9.5 in just a single session, running on a mini heater so to speak. So the 7000 hand sample is quite obviously ridiculously insanely tiny

About 200k

I think it's nearer 350k (about), but it depends how quickly you drop the 10 buy-ins.

e.g. If you won at 5bb/100 for 100,000 hands, then you'd have made a profit of 50 buyins.
If you somehow lost 10BI in your next 10,000 hands, that's 20% of your profit up in smoke, and your winrate has dropped to 40 BI per 110,000 hands = 3.6bb/100.
If it took you 50,000 hands to lose 10BI, then your overall winrate would be 40/150,000 = 2.7bb/100. It's kind of counter-intuitive, but when the downswing is quick, it doesn't drag your 'lifetime' winrate down as much as a more protracted downer.

FWIW, I won at 80bb/100 today, and got my winrate for this month up to minus 4bb/100 like a boss. (Until yesterday, I was on minus 25bb/100) #LolbadSamplesize #variance

very detailed answer! So even after 100,000 hands, a 10,000 hand negative 10 BI swing brings it from 5bb/100 to 3.6bb/100! That is still a pretty extreme change of winrate!

OP said one day so I assume pretty much instantaneously. 200k hands is 100 buy ins, losing 10 buy ins means a 10% deduction so you go from 5 to 4.5.