How big? Under Eugene Nalimov’s modification of Thompson’s original file format, the database for up to 5 pieces takes 7.05 gigabytes, compressed over 75% from the originals. The complete compressed 6-piece set fills 1.2 terabytes. That’s on the order of large data set examples given by Pankaj Agarwal in a University at Buffalo CSE Distinguished Speaker lecture last month, for NASA, Akamai, the LSS Telescope, and others. The complete 7-piece set has been estimated to be around 70TB, and its computation may finish by 2015. Nalimov’s distance-to-mate (“DTM”) files are built iteratively working backwards from checkmate positions. The newer ones (and Thompson’s originals) use distance-to-conversion (“DTC”), which means including all winning positions with fewer pieces in the ground set. Following chess endgame practice, let’s call the winning side White. The first iteration marks all positions where White can give checkmate—and/or make a winning capture—in one move. The next iteration finds all positions where Black can’t avoid the previously-marked positions (or in DTC, must make a losing capture). Note that this is only a subset of the usual graph neighborhood of the marked positions—it’s the subset of for which all out-neighbors are marked. But the next iteration (with White again to move) marks all the positions which has an edge to the previously-marked Black-to-move positions. The 517-move win was the only unmarked position left (up to symmetry) in the 1,034th iteration for those particular pieces, after months of runtime