Revolution - Revolution is a free, open-source UCI chess engine derived from Stockfish. Jorge Ruiz Centelles, with credit to ChatGPT, modifies and extends the code to explore new concepts. The engine implements cutting-edge search algorithms combined with neural network evaluation. Derived from fundamental chess programming principles, Revolution analyzes positions through parallelized alpha-beta search enhanced with null-move pruning and late move reductions.
As a UCI-compliant engine, Revolution operates through standard chess interfaces without an integrated graphical interface. Users must employ compatible chess GUIs (Arena, Scid vs PC, etc.) for board visualization and move input. Consult your GUI documentation for implementation details.
Revolution 2.74-dev-250925-tsd1 what's new?
Summary
Made the tempo bonus explicitly symmetric so both colors receive the same +10 centipawn incentive for having the move, removing the previous color-dependent bias in evaluations
Refreshed CLI utilities so their default paths and display names use the new binary identifier across tuning, perft, match, and metrics tooling.
Updated Fritz/Cutechess/Fastchess harness configurations to reference the renamed executable and show the new identifier in GUI results.
Version 2.74 introduces a new heuristic king_file_exposurethat adds up to 6 points per column and applies a weight dependent on the weighted material before combining it with the NNUE evaluation. This adjustment can shift the result by tens of centipeons, so any miscalibration aggressively penalizes open positions compared to 2.40, which delegated this evaluation to the network.
Correction history now combines all histories into a single value, scales it by a depth-dependent factor, and reduces it by an additional 22% when it detects prior experience. It also king_file_exposurereduces or amplifies bonuses before backpropagating them, which can negate learned corrections or magnify punishments when the king is exposed.
The experience subsystem has been rewritten to support multiple formats, convert files .bin, and merge duplicate entries, but now deletes and recreates the table on every load, stores BrainLearn headers, and shares a experienceAvailableglobal that alters history weighting even across threads that haven’t consumed data; if the file is misaligned or oversized, the table becomes empty and the experience is disabled mid-search.
SPRT
Elo: 8.37 +/- 15.06, nElo: 16.29 +/- 29.30 LOS: 86.20 %, DrawRatio: 52.59 %, PairsRatio: 1.25 Games: 540, Wins: 137, Losses: 124, Draws: 279, Points: 276.5 (51.20 %) Ptnml(0-2): [3, 54, 142, 69, 2], WL/DD Ratio: 0.82 LLR: 0.50 (17.0%) (-2.94, 2.94) [0.00, 10.00]
Elo: 4.96 +/- 26.67, nElo: 10.73 +/- 57.55 LOS: 64.26 %, DrawRatio: 61.43 %, PairsRatio: 1.08 Games: 140, Wins: 32, Losses: 30, Draws: 78, Points: 71.0 (50.71 %) Ptnml(0-2): [0, 13, 43, 13, 1], WL/DD Ratio: 0.65 LLR: 0.07 (2.3%) (-2.94, 2.94) [0.00, 10.00]
Elo: -0.00 +/- 15.39, nElo: 0.00 +/- 29.86 LOS: 50.00 %, DrawRatio: 52.69 %, PairsRatio: 0.98 Games: 520, Wins: 133, Losses: 133, Draws: 254, Points: 260.0 (50.00 %) Ptnml(0-2): [2, 60, 137, 58, 3], WL/DD Ratio: 1.01 LLR: -0.22 (-7.3%) (-2.94, 2.94) [0.00, 10.00]
Revolution 2.74-dev-250925-tsd1 - download mirror
Comments
Post a Comment