For fans of opposite castling who want firecracker games:

[Dann Corbit]

This 7-zip archive:

Contains 13,598 opposite castled positions that came from real games. (Possibly a few minor exceptions from study positions in the main table).

All of the positions have been analyzed to 30 ply or more and all have an evaluation between .5 and 1.0 pawns.
Now for the down-side. Since they came from real games, there are many clumps of positions which are from the same game.
Also, I do not have evaluations or other decoration because some of them have hundreds of records for a single position (all of which fit the criteria).
I collected the data with these two SQL queries:

This one collects from my contest games table:

Code: Select all

select distinct Epd from dbo.ContestGames where Epd like '2k%K1 % - -' and abs(ce) between 50 and 100 and acd >= 30
union 
select distinct Epd from dbo.ContestGames where Epd like '%k1/%/%/%/%/%/%/2K% - -'  and abs(ce) between 50 and 100 and acd >= 30

And this one collects from my main study table:

Code: Select all

select distinct Epd from Epd where Epd like '2k%K1 % - -' and abs(ce) between 50 and 100 and acd >= 30
union 
select distinct Epd from Epd where Epd like '%k1/%/%/%/%/%/%/2K% - -'  and abs(ce) between 50 and 100 and acd >= 30

These tables do not have the same shape, but both do contain analysis. It is just that the contest games table has lots and lots of multiples.

[Dann Corbit]

Another difficulty with the data is that the engine that analyzed can be any engine.
Some engines (like certain versions of ExChess) have extreme depth due to over-pruning.
Some engines may have an inaccurate eval.