I have plenty of endgames that engine evaluates like +1, even +2, not changing much for like 50-100 halfmoves, repeating, playing without plan. Defending side seems to be always just in time on right corresponding squares to hold. The evaluations are from 2m+2s engine blitz games, starting from low material endgame position, plenty of time to make good calculations in bishop/rook/knight endgame I think. How certain can I be such positions are really draws? Is it possible that engines would miss winning line with so much time more than exceptionally? Not in study type positions but in normal practical endgames. Is for example sightseeing tour with king from g4 to g8 through a4 in opposite colored bishop 3 vs. 1 still beyond horizon of engines? They often don't use easy plan I would try to win the game and I wonder if they see all the corresponding drawing setups all over the board or if they are still living in a box in technical endgames.
Chess engines are notoriously optimistic in endgame positions. I'm sure Stockfish would give you a healthy margin for many theoretical drawn rook endgames, where the defending side is a pawn down.
You can't just look at the evaluation score. You should try to interpret the engine line, and use your common sense. If you have something like +8, most likely it's a win but +1 isn't always sufficient.
Engines are not perfect, and they may not evaluate a long-term strategic position correctly. Opposite colour bishops are good example.
Do you know you can use Syzygy 6-men tablebase to boost your engine endgame performance. Most serious engine users have tablebase.
In the endgame engines will give one of three kinds of evaluation according to what they "see".
1) #31 - mate in 31. Material balance is irrelevant.
2) 0.0 - they see a forced draw. Material balance is irrelevant.
3) +2 or -1 or whatever - they don't see mate and they don't see a forced draw. They do see a material imbalance. Any concrete result is outside their move threshold and realistically any one of three results is possible. If the evaluation doesn't change much over a prolonged number of moves it shows that one side has a material advantage but can't use it to win as far as the computer can "see".
Remember computers don't have judgement. They are just calculating machines. They can only tell you what they calculate.