By design. Markets like Over/Under a goal line and Both Teams To Score have exactly two outcomes, and one of them must happen. If the model assesses Over 2.5 at 58%, Under 2.5 is 42% by construction. The two sides are complements of each other, and the model enforces that relationship on every two-outcome market it prices at full time.
The rule keeps each market internally consistent. Both sides derive from one coherent probability, so the fair odds for Over and for Under can never contradict each other. Without it, a system could assess Over at 58% and Under at 47%, publish fair odds from each, and hand you two readings of the same match that cannot both be true. Enforcing the complement also keeps the +EV figure on each side honest: any value shown reflects a real gap between the model’s view and the bookmaker’s price, not an accounting inconsistency inside the model.
It doubles as a reading aid. When the model prices Over as likely, it prices Under as unlikely to match, and the question worth asking becomes which side of that split the bookmaker’s prices disagree with.