Structural Factor Framework for Gomoku Evaluation: Falsifiable Hypotheses, Engine Blind-Spot Characterisation, and a Pre-Registered Verification Protocol
This paper formalises a set of experience-based judgements about Gomoku, accumulated through long-term independent practice, into a structural factor evaluation framework.本文将通过长期独立实践积累的一组五子棋经验判断,形式化为一个结构因子评估框架。The framework publishes its factor list and aggregation structure while sealing its numerical weights.该框架公开发布其因子清单与聚合结构,同时将数值权重保密。Three falsifiable hypotheses are derived: (H1) the evaluation gap between the best move and the second-best move is often small in master-level play; (H2) mainstream engines exhibit systematic bias on long-jump and distant-defence structures, and the bias grows toward the endgame; (H3) the marginal value of a live two decays monotonically with game phase.由此引出三条可证伪假设:(H1)大师级对局中最优手与次优手的评估差距常呈“平坦”分布;(H2)主流引擎对远跳与远防结构存在系统性偏差,且该偏差随对局向残局推进而放大;(H3)活二的边际价值随对局阶段单调衰减。A pre-registration-style verification protocol is specified before any testing begins.本文在任何测试开始之前给出预注册风格的验证协议。This paper explicitly does not claim that a human using this framework can defeat top engines, that any world champion has played incorrectly, or that forbidden-move Gomoku admits a controllable outcome for either side.本文明确不声称:人类使用本框架可击败顶级引擎、任何世界冠军存在“下错”、或带禁手五子棋可任意操控任一方胜负。Instead, it argues that the only credible form of a "bombshell" in this field is a set of reproducible artifacts: an engine blind-spot challenge suite, a machine-verified endgame corpus, a counterexample list published before successes, and pre-registered tests.本文主张:在本领域能够可信构成的“炸弹”唯一形式,是一组可复现工件——引擎盲区挑战套件、机器验证残局库、先于成功结果发布的反例清单,以及预注册测试。
Authors
- 湖南耒阳 曾乐
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24409320
- Primary Topic
- Artificial Intelligence in Games
- Type
- article
- Field-Weighted Citation Impact
- 0.00