Latent Dynamics · 2026-11-10

How Many Exposures Does a World Model Need? Learning Unseen Game Mechanics at Test Time

Ku Onoda, Haruto Kawano, Shota Takashiro, Kohei Hayashi

arXiv:2611.12324PDF

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

The authors study how many “exposures” (occurrences) a pretrained world model needs at test time to learn unseen game mechanics. They introduce MechEval to label 69 recurring mechanics across 20 ARC-AGI-3 games and score per-mechanic predictions by exposure count. Their testbed WARC updates all parameters after each transition; WARC-Agent then improves closed-loop efficiency by using WARC for action ranking and test-time training.

Why it matters

MechEval provides a protocol to measure adaptation of world-model predictions independent of exploration/return by scoring the effects of each mechanic at equal exposure counts. This directly answers how quickly a model learns the underlying rules of a new environment, and they test whether full test-time training improves mechanics learning and closed-loop action efficiency.

Method

  • Introduce MechEval: instrument ARC-AGI-3 game programs to log which mechanics fire and which cells they change; score per-mechanic prediction accuracy over exposure k.
  • Build WARC: a 35M-parameter vision transformer pretrained on other games that predicts next frames and updates all parameters after each transition (with replay and augmentations).
  • Evaluate closed-loop with WARC-Agent: pair WARC with a fixed graph explorer that ranks actions by novelty of predicted outcomes; measure RHAE on public games.

Limitation

MechEval keeps a mechanic only if it fires at least three times in the median recording that contains it.

Abstract (from arXiv)

World models enable agents to plan and learn from imagined experience, but in an unfamiliar environment they must learn its mechanics from limited interaction. A mechanic is a rule for how an action or an interaction changes part of the environment. Each occurrence of a mechanic that the model observes is one exposure. We ask how many exposures a world model needs to predict the effects of a new environment's mechanics. We address this question on ARC-AGI-3, a benchmark of interactive games defined by their programs. We introduce MechEval, which labels 69 recurring mechanics across 20 games from these programs. It scores each prediction on the cells the mechanic changes at each exposure. As a test bed, we build WARC (World model for ARC-AGI-3), a 35M-parameter vision transformer that is pretrained on other games and updates all of its parameters after each transition. On held-out games, WARC learns to predict four in five mechanics with at least 80% accuracy on the cells they change and keeps that accuracy to the end of the recordings. Half of them reach that accuracy within about 80 exposures, in a single play of the game, and without updates only one in seven does at all. In closed-loop play on the public games, WARC-Agent, which pairs WARC with a fixed graph explorer, triples the explorer's official efficiency score from 0.15 to 0.46, mostly through test-time training. On the hidden games of the Kaggle leaderboard, our entry scores 0.20.

Related papers

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First announced by arXiv: 2026-11-10 · Indexed by Pub-AI: 2026-11-10 UTC

preprint

How Many Exposures Does a World Model Need? Learning Unseen Game Mechanics at Test Time

Ku Onoda, Haruto Kawano, Shota Takashiro, Kohei Hayashi

First announced by arXiv: 2026-11-10 · Original record ↗

arxiv
2611.12324
Explicit source versions (1)
  • v1 · Date not supplied
  1. Preprint available

    Source date: 2026-11-10 · First detected: 2026-11-10 UTC · Historical baseline

    Source evidence (1)
    • Official evidence ↗ · RSS guid=oai:arXiv.org:2611.12324v1; arxiv:announce_type=new; item pubDate (official metadata)
Abstract · arxiv

World models enable agents to plan and learn from imagined experience, but in an unfamiliar environment they must learn its mechanics from limited interaction. A mechanic is a rule for how an action or an interaction changes part of the environment. Each occurrence of a mechanic that the model observes is one exposure. We ask how many exposures a world model needs to predict the effects of a new environment's mechanics. We address this question on ARC-AGI-3, a benchmark of interactive games defined by their programs. We introduce MechEval, which labels 69 recurring mechanics across 20 games from these programs. It scores each prediction on the cells the mechanic changes at each exposure. As a test bed, we build WARC (World model for ARC-AGI-3), a 35M-parameter vision transformer that is pretrained on other games and updates all of its parameters after each transition. On held-out games, WARC learns to predict four in five mechanics with at least 80% accuracy on the cells they change and keeps that accuracy to the end of the recordings. Half of them reach that accuracy within about 80 exposures, in a single play of the game, and without updates only one in seven does at all. In closed-loop play on the public games, WARC-Agent, which pairs WARC with a fixed graph explorer, triples the explorer's official efficiency score from 0.15 to 0.46, mostly through test-time training. On the hidden games of the Kaggle leaderboard, our entry scores 0.20.