The standard unit for a poker win rate: big blinds won per hundred hands. Chips and currency totals are not comparable across stakes or session lengths, so results are normalised — a player up 250 big blinds over 5,000 hands has a win rate of 5 bb/100.
The unit exists to be compared, and the comparison is usually humbling. Solid winners in most online games run somewhere in the low single digits; a sustained 10 bb/100 in a full-ring game is exceptional rather than typical. Any figure well outside that band is almost always a sample-size artefact rather than a discovery.
The reason is variance, and its scale is easy to underestimate. Standard deviation in no-limit hold'em runs around 100 bb per 100 hands — roughly twenty times a good win rate. Run the arithmetic and the conclusion is uncomfortable: even after 100,000 hands the 95% confidence interval around a measured win rate is still about plus or minus 6 bb/100. A player who has genuinely earned 5 bb/100 over that sample cannot yet distinguish themselves statistically from a break-even player.
The practical consequence is that a win rate is a slow instrument, and almost nothing you want to know arrives fast enough through it. Ten thousand hands says very little. A downswing of thirty buy-ins is a normal event, not evidence of a leak. Decisions about whether a specific adjustment is working have to be made by examining the decisions themselves, because the results will not answer in time.