597 moves. The longest DTZ ever computed.
The longest 7-piece win takes 517 moves to its first zeroing move. At eight pieces, with a pair of blocked pawns, the longest win reaches 597 moves, far beyond the 50-move limit.
An integrated pipeline builds and probes every supported metric up to 8 pieces, starting with the 1.3% that fits on a workstation today.
Full 8-piece coverage is seven doublings away.
Sizes are projected from existing WDL and DTZ tables.
Every 8-piece endgame with at least one pair of opposing pawns blocking each other on a file. This partial domain makes up a meaningful slice of the 8-piece space, now with exact results at every halfmove clock.
Tablebase statistics597 moves. The longest DTZ ever computed.
The longest 7-piece win takes 517 moves to its first zeroing move. At eight pieces, with a pair of blocked pawns, the longest win reaches 597 moves, far beyond the 50-move limit.
The written proof covers the full path from generation to probing. As coverage expands, the same algorithms and correctness guarantees apply.
Read the proofCorrectnessA proof shows the method is correct. ChessTB ships the tools for independent verification.
Clear names, focused modules and a standalone probing library give your agent a direct path into the code. The documented formats and written proof explain how the pieces fit together. Point it here and tell it what you want to build.
Add ChessTB endgame tablebase probing to this project. The file format is specified in README.md, the probing API in src/probe/probe.h, and the correctness proof in CORRECTNESS.md, all at github.com/noobpwnftw/chesstb.