Expected value is the probability-weighted outcome minus cost. If your probability estimate is 55% against a price implying 52%, the difference is the estimated edge — before fees, and only as good as the estimate itself.
With a small theoretical edge, a large number of observations is needed before results can be distinguished from noise. Short-run results are dominated by variance.
Estimated probabilities carry error. Treating a point estimate as certain is the most common analytical failure in this area — which is also why this site publishes no confidence scores it cannot substantiate.