Probability is not certainty
A true 55% opportunity loses roughly 45 times in every 100 trials over a very large sample. Any short sequence can contain several losses even when the underlying estimate is sound.
Judge the decision and result separately
Win/loss grading determines the actual record and units. CLV evaluates whether the locked price compared favorably with the closing market. Calibration tests whether stated probabilities match observed frequencies over time. Each answers a different question.
Losses still require review
Calling a wager “good” cannot become an excuse. We review whether the input data was complete, whether the price was captured correctly and whether the market repeatedly identified information our model missed. Model changes require chronological evidence rather than a reaction to one result.