Rodent V 1.1

This is an unusually fast second release, but we have a few good reasons to do it:

• two solid algorithm improvements (new history gravity formula and scaling NNUE eval against remaining material)
• discovering that a NNUE adjusted by training on Anand's games is better than default
• considerably reducing memory footprint
• adding UCI_Elo option, so that you can play against a weaker versions of personalities
• fixing NNUE access for personalities
• adding user manual

Rodent gained about 70 Elo in a week, and should be at 3470 CCRL at short time control.

Test against Rodent 1.0:

--------------------------------------------------
Results of RodentTest vs RodentBase (40+0.4, 1t, 32MB, UHO_Lichess_4852_v1.epd):
Elo: 78.45 +/- 15.33, nElo: 144.16 +/- 27.22
LOS: 100.00 %, DrawRatio: 42.17 %, PairsRatio: 4.84
Games: 626, Wins: 231, Losses: 92, Draws: 303, Points: 382.5 (61.10 %)
Ptnml(0-2): [0, 31, 132, 130, 20], WL/DD Ratio: 0.86
LLR: 2.95 (100.1%) (-2.94, 2.94) [0.00, 5.00]
--------------------------------------------------
SPRT ([0.00, 5.00]) completed - H1 was accepted

---

Rodent Asylum - Test of NNUE personalities

Rank Name                          Elo     +/-   Games   Score    Draw 
   1 Rodent V 1.1 Anand            100       9    3000   64.1%   44.0% 
   2 Rodent V 1.1 Nakamura         -26       9    3000   46.3%   48.3% 
   3 Rodent V 1.1                  -35       9    3000   45.1%   49.2% 
   4 Rodent V 1.1 Tal              -38       9    3000   44.6%   47.9% 

Rodent V 1.0

We are proud to announce Rodent V

https://github.com/nescitus/Rodent-V

Developed since March 2026, it can use both NNUE and HCE eval and impersonate various players true to Rodent series spirit. At its strongest setting it should perform at 3400 CCRL Elo level. Lazy SMP gains about 130 Elo at 4 cores.

Sample result

Score of Rodent-V 1.0 vs stash-35.0-windows-x86_64-modern: 209 - 58 - 133 [0.689]
...      Rodent-V 1.0 playing White: 115 - 24 - 61  [0.728] 200
...      Rodent-V 1.0 playing Black: 94 - 34 - 72  [0.650] 200
...      White vs Black: 149 - 118 - 133  [0.539] 400
Elo difference: 138.0 +/- 28.8, LOS: 100.0 %, DrawRatio: 33.3 %
SPRT: llr 0 (0.0%), lbound -inf, ubound inf
400 of 400 games finished.

Usage

Rodent is meant to be used within a chess GUI (Arena, CuteChess, Banksia etc.) As a default, it is supplied with an embeded net, so it works out of the box. For more options, after downloading, create a folder for the engine, and place there three subfolders - nets, books and personalities - downloaded from the GitHub site of the project. That way you will get access to non-default settings.

Options

Normally, upon receiving uci command, Rodent will display these UCI options:

option name Hash type spin default 16 min 1 max 4096
option name Clear Hash type button
option name Threads type spin default 1 min 1 max 256
option name PersonalityFile type string default personalities/rodent.txt

All the personalities magic will be hidden within personality files. But if you start the engine using command:

rodent_v alloptions

path to personality file will disappear and several UCI options will be displayed instead. A text file explaining them all will be provided at a later date.

Developement

As tradition demands, Rodent V begun with the code of Sungorus 1.4, this time through a Go port by Naman Thanki. Writing a chess engine in Go proved surprisingly smooth. Expected performance problems appeared only during NNUE implementation, and adding some assembly fixed them.

The search is fairly modern, although some techniques are still missing. History pruning never worked for us and SEE pruning is not yet implemented. the engine is also very cautious about not pruning moves that give check. There is a lot of scope for future improvements, and we intend future developement.

We still have many ideas how to improve our engine, both in strength and usability, but current state of our engine warrants the initial release.

Weakening

From user experience perspective, the most important missing feature is weakening mechanism. For now, it can be emulated by setting nodesLimit in personality files. 65536 nodes per move seemed enough to reach Fruit 2.1 level in blitz, at least with more balanced personalities.

Have fun with Rodent!

Naman Thanki, Pawel Koziol