Since you’re here...

We hope you will consider supporting us today. We need your support to continue to exist, because good entries are more and more work time. Every reader contribution, however big or small, is so valuable. Support "Chess Engines Diary" even a small amount– and it only takes a minute. Thank you.
============================== My email: jotes@go2.pl



Revolution 2.81 021025 - UCI chess engine (based on Stockfish)

on


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 2.81 021025 what's new?

Executive Summary.
Across 6,980 games of head-to-head testing under identical conditions (TC 10+0.1, 1 thread, 32 MB hash, opening suite UHO_2024_8mvs_+085_+094), Revolution 2.81 (build 021025) delivers an aggregate gain of +15.3 Elo ± 4.4 over version 2.70, corresponding to an overall score of roughly 52.0%. This is a small-to-moderate, reproducible improvement that is suitable for production deployments and continuous-integration gauntlets.

Methodology note (friendly but rigorous): the overall Elo is computed by inverse-variance weighting of per-run Elo estimates (the reported "±" terms), which is standard when combining multiple independent gauntlets. It naturally down-weights smaller or noisier runs and up-weights larger, tighter ones.

Changelog (high-level)
Strength: ~+15 Elo vs Revolution 2.70 at 10+0.1 / 1t / 32 MB (aggregate across all gauntlets).
Stability: Consistent advantage across multiple long matches; LOS values frequently ≥ 95% on larger runs.
Testing hygiene: Paired openings (two games per line with color swap) and fixed random seed for reproducibility.
Reproducibility
Opening suite: UHO_2024_8mvs_+085_+094.pgn
Time control: 10+0.1 Threads: 1 Hash: 32 MB
Pairing: -games 2 (color-reversed pairs), order=random, fixed seed.
Launcher: FastChess


Keep experience/learning off (or the same .exp file in read-only) for both engines.
Use the same NNUE/weights for both binaries unless you’re explicitly testing a net change.
Avoid stopping mid-round with -concurrency 2 to preserve clean pairing statistics (target PairsRatio ≈ 1.00).
For live metrics, tail the log (PowerShell):
Get-Content "…\revo_2_81_vs_2_70.log" -Wait | Select-String -Pattern 'Elo|LLR|LOS|SPRT|rating'
What users can expect
Practical strength: A reliable ~+15 Elo uplift at the tested settings/book, translating to a small but tangible edge in self-play, analysis, and large-batch testing.
Portability: While Elo is control- and book-dependent, the consistency across long runs suggests the improvement generalizes to nearby settings.
Statistical caution: Historical runs occasionally showed PairsRatio > 1.1 (incomplete color-reversed pairs), which reduces SPRT efficiency. The aggregate estimate above mitigates this by weighting runs by their uncertainty. New runs with strict pairing should converge faster.
TL;DR
Revolution 2.81 (021025) delivers a ~+15 Elo gain over Revolution 2.70 across 6,980 games at 10+0.1, 1 thread, 32 MB, using the UHO_2024_8mvs_+085_+094 suite. Results are consistent across large gauntlets, confirming a steady, production-ready improvement


Revolution 2.81 021025 - download   mirror







Comments

Post a Comment