Identifying fortress draws

[h.g.muller]

I added an easy trick to my EGT generator for getting a better impression of how many fortress draws a given end-game contains. Normally this would be masked in the statistics by the much larger number of draws due to grabbing an unprotected opponent piece (usually on the first move), changing the material balance.

What I do is instead of normal seeding the EGT with wins and losses from the successor EGTs, I also count all draws there as wins for the weak side. I do this after the normal generation (based on the win/loss seeding) has finished, giving the draw seeds a DTx larger than any that occurred so far, and then continue generation from those. (In practice my generator is one-sided, distinguishing only white wins from white non-wins, so what I do is generate the reverse end-game seeded in this special way from the normal successor end-games, where all seed positions are color-flipped, and the non-won positions (draws + losses) are then used as win seeds in so far these positions were not already won in the regular way.)

This identifies the positions where black can force conversion to a drawn position in a simplified game. In end-games that are generally won to white this in most cases has to happen very quickly, based on unfortunate initial positioning of the white pieces; often more than 90% happens in the first move. The remaining drawn positions then give the fortress draws within the end-game, where black can prevent progress without getting checkmated.

[Drennon]

The game needs to really be simplified for this to work, no?