There is an interesting position in the Traxler Counterattack where the engine is completely wrong.
Here an engine will recommend c3 at low depth, which is completely losing, although it is not very obvious why. At higher depth however an engine will recommend d6, the correct move.
Engines aren't that great at opening analysis, especially at low depths. You'd be better off using a database and going through all the popular moves in your chosen opening in that. You'll get a better sense of what moves have been played before and why they work or why they don't
the right approach is a combination of engine and database analysis. They are a lot of positions an eval can say one thing but not quite capture the difficulties even master level players may have in such positions., and the database data would reveal that.
but blind use of a database is misleading too, some lines which are outright refuted will still show up in database numbers, (but if you pay attention to top games, may practically dissapear at the highest level pass a certain date). This is especially true for suspicious speculative gambits.
#1
"at how much depth I should be looking at from the engine." ++ The more, the better. That is why top grandmasters pay for cloud engines instead of desktops.
Research shows that chess is most complicated at 26 men i.e. after 6 captures i.e. 3 trades.
That is why engines cannot evaluate properly at more than 26 men: there are too many possibilities and the horizon effect makes an effective evaluation impossible. Thus above 26 men use a data base of games.
I’m trying to use a chess engine to evaluate some less-known lines in openings. One problem I can see with this is that the engine evaluates the quality of the position after, let’s say 30 moves. So, let’s say I pick a move with the best score. However, that score might be based on a specific sequence of my next 15 moves. It is not likely that I would be able to follow that path exactly (assuming for simplicity that the opponent does pick the computer moves). If I deviate from that path, the results might potentially be disastrous. So, from the practical perspective, I would prefer a move that gives me good positions in the future under a wider range of my moves. So, I guess I also need to pay attention to moves that are generally strategically sound.
Also, in that context, it is interesting at how much depth I should be looking at from the engine. For example, if a move is the best at depth 24 (let’s say score 0), but then becomes -0.3 behind other moves at depth 30, is this move good enough for me? As I play this opening in the actual game, I am not likely to follow that sequence planned by the computer from depth 24 to 30.
Another possibility is that those evaluations at high depth are indeed mostly reasonable assessments of the strategic value of a given move. I’m just wondering, if anyone has any experience or thoughts on this.