From looking at the block stats on etherscan over the last 20 days, it looks like there are approximately:
475-525 single depth reorgs per day (uncle blocks)
1-2 reorgs with depth of two per day
I didn't see a single depth of three or more reorg
These numbers will probably go out of date soon. Only a few months ago there were days with 2,000 uncle blocks a ...
Just to follow on from Daniel's answer, here's a crappy Python scraper to scrape Etherscan and count the number of ephemeral forks of different lengths.
Forks of length 1: 6526
Forks of length 2: 164
Forks of length -: 85699
(Note: I think the - values are from older blocks where the depth data wasn't being recorded by Etherscan... so of limited ...
"Best" depends on the goals.
Blockchain provides "finality" through a data structure and a process that makes it possible to reason about finality. That is a deliberately vague description because there are many variations on the structure, the process and the idea of confirmation.
BFT uses validators. They are privileged and they could (...
Just to add to @Daniel's great answer: the amount of necessary confirmations depends on the needed amount of surety. How sure does the party need to be that the block won't get reverted?
As noted, single block reorgs happen all the time. Anything beyond that reduces the chances of revert drastically, per every extra depth.
Basically if you'd have data ...
You can consider a transaction 100% final, if there is an upper limit to network delay between nodes and 51% of validators are honest and have signed your block.
If you can't guarantee 51% honest validators, you need 100% of validators signing your block, 66% of them being selfish and 1 being honest.
It's in reach as much as it is theoretically possible: it can provide theoretical finality guarantees if more than 2/3 of participants are honest. These are the safety guarantees that Byzantine Fault Tolerant systems like Casper can offer, so much of the work is on attempting to discourage byzantine behaviour and minimising the damage when finality fails (e....