Chapter 2: Selected Technical Posts from the Qizhong Forum (棋中论坛)

2.1 Technical Post: Adaptation of Null Move Pruning in Chinese Chess

The discussion on the Qizhong Forum about Null Move Pruning is an important case of technical exchange in Chinese Chess engines. Below is a compilation of the forum discussion:

Topic: R-value Selection for Null Move in Chinese Chess

The choice of R-value (the depth reduction value for null move pruning) has a significant impact on the effectiveness of null move pruning:

• R=2: Aggressive pruning, fast search speed, but may miss critical moves
• R=3: Moderate pruning, recommended default value
• R=4 or higher: Overly aggressive pruning, not recommended for most positions

In Chinese Chess, the adaptation of null move pruning needs to consider:

1. King face-to-face: Null move pruning must be disabled in positions where the Kings face each other
2. Endgame phase: The aggressiveness of null move pruning needs to be reduced in the endgame (because each move carries greater value)
3. Check positions: Null move pruning cannot be used when in check

2.2 Technical Post: Evaluation Function Tuning Strategies

Discussion on the Qizhong Forum about evaluation function tuning:

The methods for tuning evaluation functions were a long-term topic of discussion in the community. The main methods include:

1. Trial and error: Manually adjust parameters, then verify the effect through self-play testing
2. Data-driven: Use large-scale game record data to optimize parameters through linear regression or more complex statistical methods
3. Comparison: Compare evaluation results of different parameter settings on specific positions

In Chinese Chess evaluation functions, the following parameters require focused tuning:

• Material values: Base values of Rook, Knight, Cannon, Advisor, Bishop, Pawn
• Positional values: Score for each piece at each position on the board
• Pattern values: Adjustment scores for specific patterns (such as exposed Cannon, trapped Knight, shoulder Cannon, etc.)

Chapter 3: The Roadmap of Chinese Chess Engine Strength Evolution

3.1 1985-1995: Early Stage

Playing strength characteristics: Beginner amateur (approximately 1200-1600 ELO)

Representative engines: Zun Zu / Chess Master (1989), Xiangqi Master 3 (象棋大师3) (1992)

Technical limitations:

• Search depth 4-7 plies (Zun Zu) to 6-8 plies (Xiangqi Master 3)
• Evaluation function based on simple material values and basic positional weights
• No transposition table, or only a very simple one
• No advanced pruning such as Null Move or LMR
• Single-threaded execution

Relationship with human opponents: Could defeat amateur beginner and intermediate enthusiasts, but easily lost to advanced amateurs and city-level players.

3.2 1995-2005: Intermediate Stage

Playing strength characteristics: Advanced amateur to city level (approximately 1700-2100 ELO)

Representative engines: Chess1 / Qi Yin (棋隐) (1997), Meng Ru Shen Ji (梦入神机) (2004), Xiangqi Qibing (象棋奇兵) (2005)

Technical improvements:

• Search depth increased to 8-10 plies
• Evaluation function added positional values and piece activity evaluation
• Introduction of transposition tables
• Began experimenting with basic pruning techniques (such as simple null move pruning)
• Still primarily single-threaded

Relationship with human opponents: Could consistently defeat players below provincial level, but still had a clear gap when facing National Masters.

3.3 2005-2015: Advanced Stage

Playing strength characteristics: National Master to surpassing Grandmaster (approximately 2200-2800+ ELO)

Representative engines: Xiangqi Cyclone (象棋旋风) (2005-2008), Chess Heaven / Qi Tian Da Sheng (棋天大圣) (2006), Intella / Yitian Xiangqi (倚天象棋) (2008), Xiangqi Mingshou (象棋名手) (2009-2014)

Technical improvements:

• Search algorithm fully upgraded to PVS
• Multi-threaded parallel search (2-4 cores)
• Distributed computing support (Mingshou)
• Highly refined evaluation function (hundreds of features)
• Comprehensive pruning techniques (Null Move, LMR, SEE, Futility, etc.)
• Large-scale opening book and endgame tablebase support

Relationship with human opponents: After 2006, top engines reached Grandmaster level. After 2010, engines comprehensively surpassed human players.

3.4 2022-2026: The NNUE Era

Playing strength characteristics: Far beyond human limits (>3300 ELO)

Representative engines: Pikafish (皮卡鱼) (2022-present), Orange (2024), Cyclone-nn (NNUE version)

Technical innovations:

• NNUE neural network evaluation replaces handcrafted evaluation
• Lazy SMP parallel search (high core count scalability)
• ATT++ and other new evaluation architectures
• Statistical validation based on Fishtest
• Open-source community-driven development

Relationship with human opponents: Engine strength far exceeds human players; games hold no suspense. Engines’ primary competition is against other engines.

Chapter 4: Commercial History of Chinese Chess Engines

4.1 Pricing and Licensing of Commercial Engines

The pricing model of Chinese Chess commercial engines has undergone the following evolution:

Early period (2000-2005): Primarily sold on CD media, prices ranged from tens to a hundred RMB.

Middle period (2005-2010): Shifted to serial number or hardware dongle licensing, prices ranged from several hundred to over a thousand RMB.

Late period (2010-2022): Adopted online activation, with pricing based on core count (e.g., 2-core version, 4-core version, 6-core version), prices ranging from several hundred to several thousand RMB.

Mingshou’s pricing strategy (example):

• 2-core version: approximately 200-300 RMB
• 4-core version: approximately 500-800 RMB
• 6-core version: approximately 1000-1500 RMB
• Distributed version: approximately 2000-3000 RMB (requires multiple computers)

Licensing restrictions for commercial engines:

• Usually bound to hardware (via CPU serial number or MAC address)
• Non-transferable (license can only be used on a specific machine)
• No refunds (trial versions provided for evaluation before purchase)

4.2 Market Competition Among Commercial Engines

The main participants in the Chinese Chess commercial engine market:

First tier (2005-2015): Cyclone, Mingshou, Intella

• Highest playing strength, highest price
• Targeted at professional players and serious enthusiasts

Second tier (2010-2020): Xiao Chong (小虫), Jiajia (佳佳), etc.

• Slightly lower playing strength but more affordable
• Targeted at a broader base of amateur enthusiasts

Third tier (2005-2015): Various small engines and tools

• Targeting specific needs (such as opening book creation, game record analysis, etc.)
• Primarily targeting entry-level users

Market landscape changes:

• 2005-2008: Cyclone and Intella — two powers vying for dominance
• 2009-2014: Mingshou rose and dominated for a long period
• 2015-2020: Overall contraction of the commercial engine market
• After Pikafish open-source in 2022: Fundamental disruption to the commercial engine market

4.3 Pikafish’s Impact on Commercial Engines

The emergence of Pikafish as a free and open-source engine had a tremendous impact on the Chinese Chess commercial engine market:

1. Overwhelming playing strength: Pikafish surpassed all commercial engines in playing strength
2. Price advantage: Free and open-source vs. hundreds to thousands of RMB
3. Update speed: Community-driven vs. individual/team development
4. Transparency: Open-source code reviewable vs. closed-source black box

Commercial engines’ coping strategies:

1. Cyclone released an NNUE version (Cyclone-nn), attempting to remain competitive with Pikafish
2. Some commercial engines lowered prices or switched to free
3. Some small engines ceased updates
4. The focus of engine development shifted from commercialization to servitization (such as cloud book subscriptions)

Chapter 5: International Influence of Chinese Chess Engines

5.1 International Community Interest in Chinese Chess Engines

With the open-sourcing of Pikafish and the participation of the international chess community, Chinese Chess engines began to receive broader international attention:

Stockfish community interest in Chinese Chess: Some contributors to the Stockfish community also began to pay attention to the Pikafish project. They brought experience and tools from international chess engine development.

Fairy-Stockfish’s Chinese Chess support: As a chess variant engine, Fairy-Stockfish’s Chinese Chess support provides a reference implementation of Chinese Chess rules from the international chess community.

International publication of academic papers: Daniel Tan’s paper on NNUE datasets was published at an international academic conference (IEEE Conference on Games 2024), marking the entry of Chinese Chess engine research into the international academic arena.

5.2 Relationship Between Chinese Chess Engines and AI Research

The position of Chinese Chess engines in AI research:

Teaching cases: Chinese Chess engines serve as classic cases for AI education, with high appeal and engagement among students.

Algorithm verification platform: Chinese Chess engines provide an excellent experimental platform for verifying search algorithms, evaluation function design, parallel computing strategies, etc.

Object of comparative research: The differences between Chinese Chess and international chess (rules, board, pieces, etc.) provide interesting research material for studying the generality of board game AI.

Chapter 6: Conclusion: The Fusion of Technology, Community, and Culture

The history of computer Chinese Chess game-playing is not only a history of technological evolution but also a cultural history of community collaboration and knowledge sharing.

From the lone struggle of Zun Zu winning the championship in London in 1989, to the continuous progress of Pikafish under global developer collaboration in 2026, the development of Chinese Chess engines reflects the formation process of a globalized community.

This history teaches us:

1. Openness is stronger than closure — Pikafish’s surpassing of commercial engines proves the superiority of the open-source model
2. Community is more resilient than individuals — a genius developer can create a miracle, but a community can continuously create miracles
3. Standardization is more valuable than fragmentation — standardization work such as the UCCI protocol, FEN format, and Fishtest framework form the foundation of a thriving engine ecosystem
4. Data is more important than code — in the NNUE era, the quality of training data determines the ceiling of engine performance

Looking ahead, computer Chinese Chess game-playing will continue to develop. Although top-tier engines have already far surpassed the limits of human playing strength, the charm and wisdom of Chinese Chess are far from exhausted. Every refinement of the rules, the birth of each new engine, and every brilliant game all inject new vitality into Chinese Chess.

May this report serve as a useful contribution to the historical record of computer Chinese Chess game-playing, and may future researchers build upon this foundation to continue recording and disseminating the story of this unique knowledge community.

(The End)

---

Volume X: Interaction Between Chinese Chess Engines and Human Players

Chapter 1: Applications of Engines in Chess Teaching

1.1 Methodology of Engine-Assisted Teaching

Chinese Chess engines play an increasingly important role in Chinese Chess teaching. Methods of engine-assisted teaching include:

Position Analysis: Coaches use engines to analyze students’ games, identifying critical misses and opportunities in the game. The engine’s precise evaluation (e.g., “win rate dropped by 15% after this move”) makes position analysis more quantitative and intuitive.

Tactical Training: Use engines to generate targeted tactical problems. Engines can precisely control the difficulty level of problems, providing suitable training for students at different levels.

Game Review: After students play against the engine, they use engine analysis to identify their weaknesses. The engine’s in-depth analysis and display of multiple variations help students understand the meaning of each move.

Opening Study: Students use engines to study opening variations and understand the strengths and weaknesses of different opening moves. Engine analysis helps students build a systematic opening knowledge system.

1.2 Engine Teaching Practices at Renowned Chess Schools

Many Chinese Chess schools and training institutions in China have incorporated engine analysis into their teaching systems. Here are some typical teaching practices:

Guangdong Chess Association’s engine teaching project: Using Pikafish for position analysis, combined with coaches’ manual explanations. In the classroom, the engine serves as an “auxiliary tool” rather than a replacement.

Beijing Children’s Palace chess training courses: Introducing engines as “playing partners,” with students analyzing games after playing against the engine. The engine is set to different difficulty levels to match students’ abilities.

1.3 Deep Integration of Engines and Game Records

The impact of engine analysis on game record research:

Traditional game record research relied on players’ manual analysis and experience-based judgment. After the introduction of engines:

1. Every game record can receive in-depth engine analysis (best moves, critical variations, etc.)
2. “Intuitive” moves in game records can be verified or refuted by engines
3. Previously unnoticed opportunities in historical games can be uncovered by engines
4. The quality of game record annotations has been greatly improved by the introduction of engine analysis