Horizon effect

The horizon effect arises because a chess engine can search only to a finite limit. That boundary is the search horizon. If a major loss occurs just beyond it, the position at the boundary may look much better than it really is.

The classic case involves an unavoidable threat. The engine finds sacrifices or maneuvers that delay the loss for a few moves and push it outside the horizon. The consequence remains inevitable, but the evaluation function no longer sees it where the search stops.

Not every missed deep tactic is a horizon effect. The defining feature is a cutoff-created illusion: postponing the problem appears to solve it because the consequence temporarily disappears from the searched tree.

A common defense is quiescence search. At the normal limit, the engine continues selected forcing moves, especially captures and, depending on the implementation, checks or promotions, until the position is less volatile. Iterative deepening, extensions that add search to special lines, and better move ordering, which tests promising moves first, reduce some cases, but no finite limit removes the risk completely.