Euphron Neo

[DeNooseDaniel]

Hello everyone,

My name is Daniel De Noose, I'm 52 years old, father of five childrens (21,18,12,10,6), and I'm a chess club player for 32 years.

I wanted to introduce you to my new, original engine (no borrowed code, my own formulas, etc.). It's still under development and is a classic alpha-beta engine with evaluation, meaning it doesn't use NNUE. It's available for 64-bit Linux and 64-bit Windows. It uses the UCI protocol. I'm currently at version 0.46.

Please feel free to share your feedback, test results, bugs report, etc., to help me improve it.

I've tried adding a few features that I thought would be useful.

I invite you to discover it:

Euphron Neo

Bonjour à tous,

Je me présente : Daniel De Noose, 52 ans, père de 5 enfants, joueur d'échecs en club depuis 32 ans.

Je voulais vous présenter mon nouveau moteur original (pas de code emprunté, formules personnelles, ...). Il est encore en plein développement et se présente comme un moteur alpha-beta à évaluation classique et donc sans NNUE. Il est disponible pour linux et pour windows en 64bits. Il utilise le protocole UCI. Actuellement j'en suis à la version 0.46.
N'hésitez pas à me donner vos avis, résultats de tests, bugs, ... afin de m'aider à le faire évoluer.
J'ai essayé de lui ajouter quelques fonctionnalités que je trouvais utiles.
Je vous laisse le découvrir :

Euphron Neo

[DeNooseDaniel]

Version 0.46 had a little bug so now enjoy the 0.47 version !

https://www.kingdomofcaissa.com

EUPHRON NEO — version 0.47

A chess engine written from scratch in C++17, developed through
individually tested iterations. A Caissa Academy project.

Author: Daniel De Noose
Language: C++17
Long-term target: CCRL ~3200-3300
Estimated strength (0.47): ~2600-2700 Elo (informal estimate, not an official CCRL rating)
Code license: original


===============================================================
CODE ORIGIN — NOTHING COPIED, ONE NAMED EXCEPTION
===============================================================

No line of Euphron's code comes from another chess engine. The
architecture, search engine, and evaluation are written entirely for this
project.

Stockfish 11 was used as a comparison reference during development — to
position Euphron's technical choices and understand published ideas. Its
code was read, never copied or compiled into the project.

The only exception, explicitly isolated: Syzygy tablebase recognition
relies on Fathom (MIT license), a third-party library dedicated to probing
these tables — not a chess engine, just a data-format reader. This is the
project's only external code dependency.


===============================================================
VERSION HISTORY
===============================================================

Developed in small iterations, each tested before being kept. Main
milestones:

0.10
Foundation: bitboard board representation, magic bitboards, legal move
generation, negamax + alpha-beta, SEE, transposition table, quiescence
search, check extension, basic UCI.

0.20
Six pruning/reduction techniques introduced: null move, RFP + futility,
razoring, LMR, aspiration windows, IIR.

0.21 - 0.27
External configuration (.cfg) loaded at startup · Kaufman adjustment ·
first seven evaluation terms (pawn structure, passed pawns, mobility, king
safety, threats, space, initiative) · storm, bishop pair, knight/rook pairs
· five speculative material imbalances · corrective release (five bugs from
an external review fixed) · four piece-square tables corrected.

0.28 - 0.30
Development of an external Texel-style tuning tool (a method close to Peter
Österlund's, using coordinate descent) · outposts, rook target files, king
infiltration, threats with real SEE calculation · singular extensions,
ScaleFactor (opposite-colored bishops), Late Move Pruning.

0.31 - 0.42
Hardening sessions: extended test coverage, config/code consistency audits,
calibration-drift fixes (material=100 anchor), two-tier lazy eval, tactical
pins, extended threats, king-danger de-escalation with no enemy queen.

0.43
Syzygy tablebase integration (via Fathom, MIT) — root and in-search
probing, 3- and 4-piece tables embedded in the executable. Should handle
Syzygy tables up to 7 pieces (untested). Open/semi-open files for the rook,
extended pins.

0.45
Tuning tool consolidation: selection by named parameter families, random
selection, selection of specific piece-square table squares, built-in help.
Fixed a missing exclusion (a Syzygy probing parameter was being tuned by
mistake).

0.46
Twenty-two individual UCI toggles (one per search technique and per
evaluation family, on top of the existing master toggle) · thinking-time
allocation based on game phase, move count, and proximity to the tablebase
zone · fixed a second missing exclusion (time-management parameters were
being tuned by mistake).

0.47 (current)
Strengthened tuning tool: minimum bound for parameters protected by a
conditional guard, automatic detector for parameters with no measurable
effect on the dataset · fixed two evaluation parameters left inactive by a
conditional guard (penalty for enemy pawns advancing toward the king,
simplified hanging-piece penalty) · two new evaluation families tied to
center structure: offensive kingside bonus when the center is locked,
queenside minority attack, static king safety reduced with no enemy queen ·
manual calibration values integrated into the hardcoded defaults.


===============================================================
SEARCH ENGINE — IMPLEMENTED TECHNIQUES
===============================================================

GENERAL FRAMEWORK
- Negamax with alpha-beta pruning, move ordering (transposition table
move, killer moves, history heuristic, counter-moves, capture history,
SEE-based capture ordering)
- Transposition table — compact entries, always-replace scheme, size
adjustable mid-game (UCI option Hash, tested up to a genuine 512 MB)
- Iterative deepening with aspiration windows
- Single-threaded search currently — the architecture is designed to not
rule out multi-threading later, but it isn't implemented yet

PRUNING AND REDUCTIONS
- Null move pruning — static or dynamic reduction formula depending on
the selected variant
- Reverse Futility Pruning (RFP) and classic futility pruning
- Razoring
- Late Move Pruning (LMP) — threshold growing with depth
- Late Move Reductions (LMR) — several formulas available, including a
custom log-log formula
- Internal Iterative Reduction (IIR)
- Each technique above now has its own UCI toggle, on top of a single
master switch that disables the whole group at once

EXTENSIONS
- Check extension
- Singular extensions — a transposition-table move judged to have no
credible alternative gets an extra ply of search

QUIESCENCE
- Quiescence search with SEE-based (Static Exchange Evaluation) capture
ordering
- Principal variation (PV) node protection against certain aggressive
cutoffs

TIME MANAGEMENT
- Standard formula (remaining time, increment, moves-to-go if announced
by the interface)
- Predictive check before starting an extra depth, reactive check every
128 nodes
- New in 0.46 — budget adapted across three zones: opening (near-full
material, within the first 12 moves), approaching the tablebase zone
(few pieces left on the board), middlegame by default

TABLEBASES
- Syzygy (3- and 4-piece) embedded in the executable via Fathom (MIT) —
root probing for move choice, and in-search probing for evaluating
positions encountered

STRATEGIC TEST SUITE (STS) MODE
- Dedicated command-line mode (--sts file.epd --depth N|--movetime ms
--output report_base), separate from the normal UCI loop — has no
effect on behavior when the executable is launched without arguments
- Standard STS weighted scoring when the EPD file provides it (0 to 100
per candidate move, rescaled to a score out of 10 per position) — not
just binary right/wrong; automatic fallback to 10/0 scoring if the file
doesn't provide the weighting
- Detailed report written incrementally (text + CSV), an interrupted test
can be resumed (--continue) without starting over
- Used to judge move-choice quality at a fixed depth or time —
complements Texel tuning, which only covers evaluation, not the
search's own behavior


===============================================================
EVALUATION FUNCTION — IMPLEMENTED TERMS
===============================================================

222 parameters in total (material, piece-square tables, positional and
search terms combined), most of them automatically tunable by the built-in
Texel tuner.

Material & piece-square tables
Value per piece type, separate middlegame/endgame piece-square table for
each of the six pieces, Kaufman adjustment (knight vs bishop value
depending on remaining pawn count).

Pawn structure
Isolated, doubled (with a "locked" variant), and backward pawns. Passed
pawns: base bonus, free, supported, connected, outside, candidate, king-
distance in the endgame.

Mobility
Linear bonus per reachable square, standard mobility area (excluding
squares occupied by a friendly piece or held by an enemy pawn), for
knight/bishop/rook/queen.

King safety
Static component (pawn shelter, open files, enemy pawn storm — reduced with
no enemy queen if the center is also locked, since 0.47) and dynamic
component (per-piece weighted attack zone, safe check squares, de-
escalation with no enemy queen).

Threats
Pawn attacks on minor/major pieces, real SEE-based threats, hanging pieces.
Extended component: king threat, restricted piece, safe pawn push, forks
targeting the queen.

Tactical pins
Differentiated bonus by type of piece pinned to its king, extra bonus if
the pinned piece is also attacked.

Space & initiative
Space: safe squares in one's own territory, weighted by game phase.
Initiative: a bonus in the direction of the already-established endgame
score, if pawns remain on both sides.

Material imbalances
Bishop pair, knight/rook pair penalties, five speculative combinations
(rook+bishop vs queen, two knights+bishop vs queen, etc.).

Outposts & files
Outposts for knight and bishop. Target files for the rook, open and semi-
open files. Enemy king infiltration by rook or queen.

ScaleFactor [PARTIAL]
Only one case covered so far: opposite-colored bishop endgames with no
other minor/major piece, reducing the endgame component of the score.

Locked center structure [NEW IN 0.47]
Detection of a genuine d/e pawn chain (diagonal, each pawn blocked by an
enemy pawn facing it). When locked: offensive bonus for kingside pawn
pushes (reduced with no queen or rook left to exploit it, replaces the
classic storm rather than adding to it), queenside minority-attack bonus
(a/b pawns alone against enemy a/b/c) if a rook is available.

INTERNAL OPTIMIZATIONS
- Two-tier lazy eval — early exit from the calculation if the
material+PST gap alone, then a second tier, already settles the
comparison against the search window
- The full calculation is still redone on every call — not yet
incremental (see "Roadmap")


===============================================================
AVAILABLE UCI OPTIONS
===============================================================

27 options in total. The 24 technique toggles (type check) can each be
switched off individually to isolate the effect of a specific technique in
a match or during analysis.

GENERAL CONFIGURATION
Hash (spin, 1-2048 MB): Transposition table size, resizable mid-session
SyzygyPath (string): Folder of additional Syzygy tables, on top of the already-embedded 3/4-piece tables
ConfigPath (string): Alternate .cfg file to load, to compare two setups without duplicating the executable

SEARCH — MASTER SWITCH + 8 INDIVIDUAL TOGGLES
UsePruning: Disables IIR, razoring, RFP, null move, LMP, futility, and LMR all at once
UseIIR: Internal Iterative Reduction, individually
UseRazoring: Razoring, individually
UseRFP: Reverse Futility Pruning, individually
UseNullMove: Null move pruning, individually
UseFutility: Classic futility pruning, individually
UseLMR: Late Move Reductions, individually
UseLMP: Late Move Pruning, individually
UseSingularExtensions: Singular extensions

EVALUATION — 15 FAMILIES
UseOutposts: Knight and bishop outposts
UseRookTargetFiles: Rook target files
UseRookOpenFiles: Open and semi-open files for the rook
UseKingInfiltration: Enemy king infiltration
UseSEEThreats: Real threat via SEE calculation
UsePawnStructure: Isolated/doubled/backward/passed pawns
UseMobility: Mobility's contribution to the score (attack maps are always still computed, reused elsewhere)
UseKingShelter: Static king safety (shelter/storm)
UseThreats: Basic threats
UseThreatsExtended: Extended threats (king threat, forks, restricted piece)
UsePins: Tactical pins
UseSpace: Space
UseInitiative: Initiative
UseChainStorm: Offensive kingside pawn-push bonus when the center is locked
UseMinorityAttack: Queenside minority attack


===============================================================
ROADMAP
===============================================================

SEARCH
- Lazy SMP: Genuine multi-threaded search sharing — the current single-threaded architecture is designed not to rule it out, but nothing is implemented yet.
- ProbCut / multi-cut: Additional pruning technique, identified but not started.
- Transposition table in quiescence: The TT is currently only used by the main search.
- SEE pruning in the move loop: Distinct from the capture ordering already in place — directly pruning moves judged losing by SEE.

PERFORMANCE
- Incremental score: Material and piece-square tables are fully recalculated on every evaluation, not maintained incrementally as moves are made/unmade.
- Per-term profiling: Measure the real cost of each evaluation block individually.

EVALUATION
- Full ScaleFactor: Only one known drawish case covered (opposite bishops) — many other documented scale-factor endgames remain to be added.
- PST overfitting mitigation: Some piece-square table values (advanced king in the middlegame) reach suspicious levels after tuning — bounding or a finer step size are being considered, not done yet.

TEST INFRASTRUCTURE
- Tournament manager: cutechess-cli or c-chess-cli — neither is in place yet.
- Opening book & SPRT: Needed to validate any strength change under full playing conditions — not built yet.

SEPARATE DERIVED PROJECTS
- Euphron NeoN: An NNUE variant considered as a separate derived project — the classical Euphron engine remains the active development.
- Idea for another engine in pure continuous self-learning: Autonomous parameter improvement through gameplay, architecturally designed but kept separate from Euphron's main codebase.



===============================================================
AT A GLANCE
===============================================================

222 config parameters
27 UCI options
0.47 current version
2 embedded Syzygy tables

[Tibono]

Hi Daniel,
you need to fix either the URLs or file names for downloading 047 version (case sensitive issue; euphron in the URL but Euphron files are stored). Thanks for sharing your work!
Eric

[DeNooseDaniel]

Hi Daniel,
you need to fix either the URLs or file names for downloading 047 version (case sensitive issue; euphron in the URL but Euphron files are stored). Thanks for sharing your work!
Eric

Thanks for this useful comment ! Changes made !

[chessica]

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Stockfish 18                   : 3553   22  21   593    79.8 %   3314   40.3 %
  2 Revolution-5.90-140626         : 3551   22  21   596    79.6 %   3314   40.1 %
  3 Cool Iris 16                   : 3551   19  19   720    79.2 %   3319   41.7 %
  4 Reckless 0.9.0                 : 3549   22  21   593    79.4 %   3314   40.8 %
  5 Stockfish 19                   : 3546   25  23   490    80.7 %   3297   38.6 %
  6 Reckless 0.10.0-dev            : 3541   25  23   490    80.2 %   3298   39.2 %
  7 Rems M-091024                  : 3532   21  20   592    77.8 %   3314   44.1 %
  8 Obsidian dev-16.08             : 3528   21  20   595    77.3 %   3315   43.0 %
  9 Coda 0.9.4                     : 3527   24  23   490    78.9 %   3298   40.2 %
 10 PlentyChess 7.0.26             : 3524   19  18   720    76.4 %   3320   43.9 %
 11 PlentyChess 8.0.0              : 3521   23  22   490    78.3 %   3298   43.1 %
 12 Tarnished v6.0 (Eternal)       : 3512   21  21   593    75.7 %   3315   42.8 %
 13 Berserk 14                     : 3508   21  20   596    75.3 %   3315   44.1 %
 14 Viridithas 20.0.0-dev          : 3485   22  22   592    72.7 %   3315   41.4 %
 15 RubiChess 20240817 (avx512)    : 3480   20  20   665    71.8 %   3317   42.6 %
 16 Igel 3.7.0 64 POPCNT AVX2      : 3470   24  23   490    72.8 %   3299   43.5 %
 17 Rebel 16.3a                    : 3459   20  20   721    68.7 %   3322   40.1 %
 18 Chess System Tal 2.06 E1019    : 3450   22  22   595    68.3 %   3317   40.8 %
 19 RubiChess 20240817 (avx2)      : 3446   20  20   705    67.1 %   3323   41.4 %
 20 Maverick NNUE 1.0a             : 3404   20  20   722    61.4 %   3323   38.5 %
 21 Alexander 7                    : 3400   22  22   592    61.7 %   3317   37.3 %
 22 Shredder Classic 6             : 3378   26  26   491    61.0 %   3301   32.4 %
 23 Willow 4.0                     : 3370   23  23   592    57.5 %   3317   32.9 %
 24 Winter 4.0 JA BMI2             : 3359   21  21   724    54.9 %   3325   30.8 %
 25 Cheng 4.48a (bundled)          : 3341   23  23   595    53.2 %   3319   29.7 %
 26 Houdidit 6.03 Pro x64          : 3327   27  27   490    53.7 %   3302   26.1 %
 27 Komodo 14.1 64-bit             : 3327   23  23   595    51.1 %   3319   29.6 %
 28 EveAnn 5.0 64-bit (bundled)    : 3273   24  25   593    43.3 %   3320   24.6 %
 29 Myrddin 0.96                   : 3255   25  25   593    40.8 %   3320   22.3 %
 30 Fable 5.1                      : 3239   29  29   490    40.8 %   3304   15.5 %
 31 IsaBB NN 4.5 (avx2, bundled)   : 3236   25  25   592    38.2 %   3320   22.3 %
 32 Sable 1.6                      : 3209   27  27   593    34.5 %   3321   15.7 %
 33 Equinox 3.30 x64mp             : 3203   28  28   490    35.8 %   3304   21.4 %
 34 Critter 1.6a 64-bit            : 3194   27  27   592    32.4 %   3321   15.9 %
 35 Houdini 1.5a x64               : 3180   30  30   494    32.5 %   3307   12.8 %
 36 OpenCritter v1.1.37            : 3169   29  30   490    31.3 %   3305   19.0 %
 37 ECE 26.8 Intuition HBuP        : 3147   31  31   490    28.7 %   3306   13.3 %
 38 monty 1.0.0                    : 3141   29  29   595    26.0 %   3323   12.6 %
 39 Crafty 25.2.1                  : 3107   27  28   722    21.6 %   3331   13.6 %
 40 Spike 1.4.2                    : 3104   32  33   490    23.8 %   3306   13.3 %
 41 Spike 1.4.1                    : 3098   31  31   592    21.5 %   3323   12.5 %
 42 Colossus Chess 2026a           : 3058   36  37   490    19.2 %   3307    9.4 %
 43 Rybka 1.0 Beta                 : 3056   35  36   490    19.1 %   3307   11.2 %
 44 Colossus 2025b                 : 3047   33  34   592    16.9 %   3324   11.8 %
 45 Euphron NEO 0.47               : 2959   44  46   490    11.7 %   3309    5.5 %
 46 Pharaon 3.5.1                  : 2895   47  49   593     7.7 %   3327    5.2 %
 47 SOS 5 for Arena                : 2835   59  62   490     6.0 %   3312    3.1 %
 48 Yace 0.99.87                   : 2815   61  64   490     5.4 %   3312    3.5 %
 49 Quark v2.35Paderborn           : 2801   64  68   490     5.0 %   3313    2.7 %
 50 PatzerUCI/FRC                  : 2746   73  33   490     3.7 %   3314    2.0 %

[DeNooseDaniel]

Code: Select all

    Program                          Elo    +   -   Games   Score   Av.Op.  Draws

  1 Stockfish 18                   : 3553   22  21   593    79.8 %   3314   40.3 %
  2 Revolution-5.90-140626         : 3551   22  21   596    79.6 %   3314   40.1 %
  3 Cool Iris 16                   : 3551   19  19   720    79.2 %   3319   41.7 %
  4 Reckless 0.9.0                 : 3549   22  21   593    79.4 %   3314   40.8 %
  5 Stockfish 19                   : 3546   25  23   490    80.7 %   3297   38.6 %
  6 Reckless 0.10.0-dev            : 3541   25  23   490    80.2 %   3298   39.2 %
  7 Rems M-091024                  : 3532   21  20   592    77.8 %   3314   44.1 %
  8 Obsidian dev-16.08             : 3528   21  20   595    77.3 %   3315   43.0 %
  9 Coda 0.9.4                     : 3527   24  23   490    78.9 %   3298   40.2 %
 10 PlentyChess 7.0.26             : 3524   19  18   720    76.4 %   3320   43.9 %
 11 PlentyChess 8.0.0              : 3521   23  22   490    78.3 %   3298   43.1 %
 12 Tarnished v6.0 (Eternal)       : 3512   21  21   593    75.7 %   3315   42.8 %
 13 Berserk 14                     : 3508   21  20   596    75.3 %   3315   44.1 %
 14 Viridithas 20.0.0-dev          : 3485   22  22   592    72.7 %   3315   41.4 %
 15 RubiChess 20240817 (avx512)    : 3480   20  20   665    71.8 %   3317   42.6 %
 16 Igel 3.7.0 64 POPCNT AVX2      : 3470   24  23   490    72.8 %   3299   43.5 %
 17 Rebel 16.3a                    : 3459   20  20   721    68.7 %   3322   40.1 %
 18 Chess System Tal 2.06 E1019    : 3450   22  22   595    68.3 %   3317   40.8 %
 19 RubiChess 20240817 (avx2)      : 3446   20  20   705    67.1 %   3323   41.4 %
 20 Maverick NNUE 1.0a             : 3404   20  20   722    61.4 %   3323   38.5 %
 21 Alexander 7                    : 3400   22  22   592    61.7 %   3317   37.3 %
 22 Shredder Classic 6             : 3378   26  26   491    61.0 %   3301   32.4 %
 23 Willow 4.0                     : 3370   23  23   592    57.5 %   3317   32.9 %
 24 Winter 4.0 JA BMI2             : 3359   21  21   724    54.9 %   3325   30.8 %
 25 Cheng 4.48a (bundled)          : 3341   23  23   595    53.2 %   3319   29.7 %
 26 Houdidit 6.03 Pro x64          : 3327   27  27   490    53.7 %   3302   26.1 %
 27 Komodo 14.1 64-bit             : 3327   23  23   595    51.1 %   3319   29.6 %
 28 EveAnn 5.0 64-bit (bundled)    : 3273   24  25   593    43.3 %   3320   24.6 %
 29 Myrddin 0.96                   : 3255   25  25   593    40.8 %   3320   22.3 %
 30 Fable 5.1                      : 3239   29  29   490    40.8 %   3304   15.5 %
 31 IsaBB NN 4.5 (avx2, bundled)   : 3236   25  25   592    38.2 %   3320   22.3 %
 32 Sable 1.6                      : 3209   27  27   593    34.5 %   3321   15.7 %
 33 Equinox 3.30 x64mp             : 3203   28  28   490    35.8 %   3304   21.4 %
 34 Critter 1.6a 64-bit            : 3194   27  27   592    32.4 %   3321   15.9 %
 35 Houdini 1.5a x64               : 3180   30  30   494    32.5 %   3307   12.8 %
 36 OpenCritter v1.1.37            : 3169   29  30   490    31.3 %   3305   19.0 %
 37 ECE 26.8 Intuition HBuP        : 3147   31  31   490    28.7 %   3306   13.3 %
 38 monty 1.0.0                    : 3141   29  29   595    26.0 %   3323   12.6 %
 39 Crafty 25.2.1                  : 3107   27  28   722    21.6 %   3331   13.6 %
 40 Spike 1.4.2                    : 3104   32  33   490    23.8 %   3306   13.3 %
 41 Spike 1.4.1                    : 3098   31  31   592    21.5 %   3323   12.5 %
 42 Colossus Chess 2026a           : 3058   36  37   490    19.2 %   3307    9.4 %
 43 Rybka 1.0 Beta                 : 3056   35  36   490    19.1 %   3307   11.2 %
 44 Colossus 2025b                 : 3047   33  34   592    16.9 %   3324   11.8 %
 45 Euphron NEO 0.47               : 2959   44  46   490    11.7 %   3309    5.5 %
 46 Pharaon 3.5.1                  : 2895   47  49   593     7.7 %   3327    5.2 %
 47 SOS 5 for Arena                : 2835   59  62   490     6.0 %   3312    3.1 %
 48 Yace 0.99.87                   : 2815   61  64   490     5.4 %   3312    3.5 %
 49 Quark v2.35Paderborn           : 2801   64  68   490     5.0 %   3313    2.7 %
 50 PatzerUCI/FRC                  : 2746   73  33   490     3.7 %   3314    2.0 %

Thanks a lot for testing !

Very nice. Which Time Control ?

[chessica]

12250/12250 games, 26% draws, time control:120+1, book:avt-2026b.pgn

[DeNooseDaniel]

12250/12250 games, 26% draws, time control:120+1, book:avt-2026b.pgn

Hi. Thanks a lot.
But where can I see the results ?