DraftFM: A FoundationModel for Day-Zero Drafting in Magic: The Gathering

📄 arXiv: 2608.19568v1 📥 PDF

作者: Brian Ward

分类: cs.LG, cs.AI

发布日期: 2026-08-20

备注: 28 pages, 6 figures, 13 tables. Sealed pre-registered forecast published ~36h before release; scoring note to follow. Also on SSRN (7257098). Code: https://github.com/brianward92/mtga Sealed forecast: https://github.com/brianward92/draftfm (tag draftfm-v1.0) Weights: https://huggingface.co/brianward92/draftfm Data: https://huggingface.co/datasets/brianward92/draftfm-frozen-eval

💡 一句话要点

提出DraftFM以解决Magic: The Gathering的日零草拟问题

🎯 匹配领域: 支柱九：具身大模型 (Embodied Foundation Models)

关键词: Magic: The Gathering 草拟策略 离散选择模型 深度学习 游戏AI

📋 核心要点

  1. 现有方法在日零阶段无法利用历史选牌数据，导致草拟策略的有效性受限。
  2. DraftFM通过将每张卡片作为775维的函数输入模型，基于当前包和已选池进行评分，解决了这一问题。
  3. 实验结果显示，DraftFM在未见扩展的选牌预测中达到了显著的顶级一致性，验证了其有效性。

📝 摘要（中文）

在Magic: The Gathering的新扩展草拟中，草拟过程在任何选牌之前开始，尽管完整的卡片列表是公开的，但监督学习模型所需的草拟日志尚不存在。本文直接研究这一日零阶段。DraftFM是一种离散选择策略，基于当前包中的可用卡片进行评分，条件是已选池和草拟状态。每张卡片作为775维的冻结函数输入模型，模型中不包含卡片身份、系列身份或使用统计，因此未见卡片的评分与熟悉卡片相同。经过149百万次人类选牌训练的网络在三个完全保留的扩展中预测选牌，达到了50.8%、60.4%和56.7%的顶级一致性。该架构在32个观察到的扩展上重新训练后，为未发布的《霍比特人》扩展生成了卡片排名，并在该扩展可草拟前约36小时发布。该排名与六位独立专家评审者的意见一致性良好。

🔬 方法详解

问题定义：论文要解决的问题是如何在Magic: The Gathering的日零阶段进行有效的草拟选择。现有方法依赖于历史选牌数据，而在日零阶段这些数据并不存在，导致草拟策略的有效性受到限制。

核心思路：DraftFM的核心思路是将每张卡片视为775维的冻结函数，利用当前包和已选池的状态进行评分，而不依赖于卡片身份或使用统计。这种设计使得模型能够在没有历史数据的情况下进行有效的预测。

技术框架：DraftFM的整体架构包括输入层、特征提取层和评分层。输入层接收775维的卡片特征，特征提取层通过固定的文本嵌入处理卡片信息，评分层则根据当前状态和已选池计算每张卡片的得分。

关键创新：DraftFM的关键创新在于其评分机制，能够在没有历史选牌数据的情况下，通过结构化特征和文本嵌入对卡片进行评分。这与传统方法依赖历史数据的方式本质上不同。

关键设计：模型包含1.6百万个参数，使用149百万次人类选牌数据进行训练。损失函数设计为优化顶级一致性，网络结构采用深度学习框架以提高预测性能。

🖼️ 关键图片

fig_0
img_1
img_2

📊 实验亮点

DraftFM在三个完全保留的扩展中达到了50.8%、60.4%和56.7%的顶级一致性，相较于随机选择的7%显著提升。此外，该模型在32个观察到的扩展上重新训练后，为未发布的《霍比特人》扩展生成了有效的卡片排名，验证了其预测能力。

🎯 应用场景

该研究的潜在应用领域包括卡牌游戏的草拟策略优化、游戏AI的智能决策支持以及其他需要在缺乏历史数据情况下进行选择的场景。DraftFM的设计理念和方法可以推广到其他类似的决策问题中，具有广泛的实际价值和未来影响。

📄 摘要（原文）

Drafting a new Magic: The Gathering expansion begins before any pick from it has been observed: the complete card list is public, but the draft logs that supervised pick models train on do not yet exist. We study this day-zero regime directly. DraftFM is a discrete-choice policy that scores exactly the cards available in the current pack, conditioned on the drafted pool and the state of the draft. Every card enters as a frozen 775-dimensional function of its public card record, structured features and a fixed text embedding, with no card identities, set identities, or usage statistics anywhere in the model, so an unseen card is scored by the same machinery as a familiar one. A 1.6-million-parameter network fitted on 149 million human picks from 29 expansions predicts held-out picks in three expansions withheld in their entirety, reaching 50.8%, 60.4%, and 56.7% top-1 agreement, where uniform chance at the opening pick is about 7%. Refitted on all 32 observed expansions, the same architecture produced a card ranking for the then-unreleased set The Hobbit, sealed with its complete cryptographic provenance and published roughly 36 hours before the set became draftable on MTG Arena. The sealed ranking agrees with six independent expert reviewers roughly as much as those reviewers agree with one another. Evaluation against realized outcomes is committed to a follow-on note, whatever it shows.