Randomness is a tricky thing, depending on the context. Consider this: if you flip a fair coin 1000 times, and end up with 600 heads and 400 tails, that would constitute a 6-sigma event. The probability of this distribution to occur is approximately 1 in 500 million, which is a near impossibility. From these perfectly valid coin-flipping statistics, one may me tempted to make a deduction that if one has a trading system that makes 600 winners out of 1000 trades in a historical backtest, then it must be some kind of "Holy Grail" system, due to its unique ability to make nearly impossible odds. This deduction would be wrong (and dangerous to one's financial health). The reasons revolve around various sorts of selection biases , which are easy to underestimate and difficult to overcome.
I'm not sure that's really true. The longer the time frame, the more it becomes about historical data and probabilities, but in the very short term, milliseconds to seconds, probability doesn't have much to do with it, it becomes more about implementation, latency, and efficiency.
Well, I'm definitely not talking about seconds. In any reasonably long period of time, it's matter of probabilities.
The best traders have a win rate from 40-60% over the long haul. The biggest HFTs like Citadel has a high win rate because they spent a fortune on infrastructure & relations - no retail trader can possibly compete in the bid-ask spread arena, in fact even other well funded HFT's can no longer compete any more since the early movers have it locked down.