# Matilda: Engine-Agnostic Search with Human Policy Guidance

arXiv: https://arxiv.org/abs/2606.25176 · HTML: https://mlsanity.com/paper/2606.25176/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 13 of its 28 references that are in the map.
- First posted on arXiv 23 Jun 2026.

## Builds on (most cited first, 13 of 13)

- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [Mastering Chess and Shogi by Self-Play with a General Reinforcement Learning Algorithm](https://mlsanity.com/paper/1712.01815/) (2017)
- [Aligning Superhuman AI with Human Behavior: Chess as a Model System](https://mlsanity.com/paper/2006.01855/) (2020)
- [Accelerating Self-Play Learning in Go](https://mlsanity.com/paper/1902.10565/) (2019)
- [Maia-2: A Unified Model for Human-AI Alignment in Chess](https://mlsanity.com/paper/2409.20553/) (2024)
- [Learning Models of Individual Behavior in Chess](https://mlsanity.com/paper/2008.10086/) (2020)
- [Detecting Individual Decision-Making Style: Exploring Behavioral Stylometry in Chess](https://mlsanity.com/paper/2208.01366/) (2022)
- [Mastering Chess with a Transformer Model](https://mlsanity.com/paper/2409.12272/) (2024)
- [Human-aligned Chess with a Bit of Search](https://mlsanity.com/paper/2410.03893/) (2024)
- [Learning to Imitate with Less: Efficient Individual Behavior Modeling in Chess](https://mlsanity.com/paper/2507.21488/) (2025)
- [Chessformer: A Unified Architecture for Chess Modeling](https://mlsanity.com/paper/2605.19091/) (2026)
- [Predicting Human Chess Moves: An AI Assisted Analysis of Chess Games Using Skill-group Specific n-gram Language Models](https://mlsanity.com/paper/2512.01880/) (2025)

## Built on by (most cited first, 1 of 1)

- [Amadeus: When Models of People Meet](https://mlsanity.com/paper/2609.35835/) (2026)

## Abstract (arXiv)

Chess engines have evolved from search-based systems optimized for strength to neural policies optimized for predicting human decisions. Existing approaches largely separate these goals: search engines achieve superhuman strength but poorly model humans, while models such as Maia-3 capture rating-conditioned behavior yet degrade at elite levels. We present Matilda, a modular residual re-ranking architecture that decouples behavioral priors from tactical search, combining a frozen human policy with an engine-agnostic search backend through a lightweight residual model. Matilda learns residual corrections over the full legal-move distribution from frozen policy context, time control, player-style embeddings, and search-derived candidate features. A zero-initialized residual head exactly recovers the frozen policy before training while optimization minimizes negative log-likelihood (NLL). Instantiated with Maia-3 and Stockfish, Matilda reduces human-move prediction NLL by 18.5% and raises top-1 accuracy from 60.1% to 66.1% on temporally held-out verified-human 3000+ Elo Lichess blitz games, with player-style embeddings contributing a further 1.8% and +0.2 percentage points (pp) respectively. Seed-paired ablations attribute these gains to search rather than additional data; the findings are replicated in Go -- decomposing expert play into recognition and verified calculation. Below 2500 Elo, where search annotations are unavailable, Matilda preserves Maia-3's performance.