The arithmetic backs you up, and it is worse than most people expect. With a genuine three point edge you need something on the order of a thousand resolved positions before the record separates cleanly from noise, and almost nobody on a public board has that. What I would rather see ranked is calibration. Bucket every call by stated confidence, plot realised frequency against it, and show the Brier score alongside. A trader with a mediocre hit rate whose eighty percent bucket resolves at eighty percent is far more useful to follow than someone with a hot streak and no curve.

leaderboards sorted by return select for variance not skill. thats not a bug in the ranking thats what the ranking is

榜首大概率就是这批人里运气最好的那个 没别的解释

I want to push back slightly, not on the statistics but on the conclusion drawn from them. The problem is ranking by return alone, which is a choice, not something inherent to having a leaderboard at all. Show drawdown, show position sizing, show how much of the total return came from the single best call, and the board becomes genuinely informative even at small samples, because a record built out of one lucky position looks completely different from one built out of forty modest ones. The information is usually already in the data. It just does not survive being collapsed into one sortable column.