Extra analysis for near alternatives

[cyanfish]

I think it would be a good idea to have the generator spend more time to look further down certain lines that are almost at the alternative threshold (assuming, of course, that it doesn't already do this?).

For example, on problem 41162, the move Ng5 is almost an alternative (see my comment for analysis). This specific case is perhaps ambiguous (because Toga jumps around a bit as it goes deeper), but I can certainly imagine that some positions might have this issue in a more concrete form.

N.B. Of course, if CPU time is an issue, then this may not be the best use of it.

[richard]

CPU time is somewhat of an issue, however there is a wider issue that no evaluation other than mate in N can be truly stable.  Looking longer at all positions will usually eventually cause jumps in the evaluation.  For example say there is a tactic that lets you go a rook up, if you let the engine run for long enough after this it should eventually find a way of leveraging the advantage into further material gain (or indeed mate). 

The generator currently only looks for 1-2 mins at each position during disambiguating (which is the part of the process where the lines in the move list come from). Normally this is long enough to see what most humans can see in a position, often it is seeing much more than that. Spending longer will lead to an improvement in a few problems, but will also likely cause more positions where the human is left scratching their head at a sequence that occurred because of a move the computer saw 20+ ply down the track.

[cyanfish]

True enough... All right then, I guess the way it works right now is the best way to do it.