A player thinking that they “have a lead” and being actually ahead are two different things, as are having a lead and actually winning.
RTS example: you might feel confident because you control more than half the space or resources and you are building everything you want efficiently and without being attacked, until you make contact with the enemy and you find out that you are threatened by a rush or that you built entirely wrong units.
Chess example: you have more pieces, but you don't see that you are falling into a trap of repeated checks and forced moves.
In some games there is no way to have a significant lead: for example, in a typical racing game the leading player can make a small mistake and be overtaken at the last curve, because the difference between a good and bad player who don't crash is a fraction of a second per lap.
What's important in most endgames is that with or without comebacks, with more ore less random outcomes, the game doesn't drag on through boring and repetitive sections and doesn't last too much. For example, battle royale PvP games often have a shrinking map to maintain a fun balance between fighting and hide and seek as players are eliminated. In chess it's usually easy, and progressively easier as pieces are captured, to force piece captures and clear the board enough to reach a simple finale. In scoring-based or mistake-dependent games there are no boring game states, only constant maximum effort.
There's an important related concept, that of “strategic instability”: a situation where someone is guaranteed to win because a standstill is impossible. For instance, while a plain multiplayer racing game could continue for an arbitrarily long time, usually needing a predetermined limit (laps, fuel, time…), a racing game where cars have front-mounted guns and dropped mines is guaranteed to end dramatically because damage can only accumulate.