# Tycho: Active Abstraction with Programmatic World Models for ARC-AGI-3

> Tycho constructs editable Python models of game state, transitions, rendering and outcomes from action/observation histories in unfamiliar ARC-AGI-3 games. It checks model predictions against recorded transitions and can use, repair or bypass these models during planning.

- Authors: Jens Lehmann; Andrei Aioanei; Sahar Vahdati
- Venue: arXiv
- Status: Preprint
- Publication / posting date: July 30, 2026
- Date source: First public arXiv submission date; version v1 was reviewed.
- First arXiv submission: July 30, 2026
- arXiv: 2607.28287
- Reviewed arXiv version: v1
- DOI: 10.48550/arXiv.2607.28287
- Sources reviewed: 2026-10-17
- Added to the catalog: 2026-10-17
- Original paper: <https://arxiv.org/abs/2607.28287>
- PDF: <https://arxiv.org/pdf/2607.28287v1>
- Code: <https://github.com/NIMI-research/Tycho>
- BibTeX: <https://awesomerea.org/papers/citations/paper-tycho.bib>
- Catalog page: <https://awesomerea.org/papers/paper-tycho/>
- Record JSON: <https://awesomerea.org/papers/paper-tycho/record.json>
- Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-tycho/index.md)

## Scope and limitations

Evaluation uses one stochastic run per configuration on 25 public games and 183 levels; five games informed harness development. Models may abstain on cells or accept bounded rendering variants. Transition replay and game scores do not establish complete engine reconstruction, correct objectives, unseen-state fidelity, unseen-game generalization or run-to-run reliability; the agent can bypass modeling.

## Why it fits this collection

Studies executable reconstruction of observed game dynamics and checks model fidelity separately from task success, providing a bounded behavioral reconstruction case.

## Publication and review notes

The arXiv posting date is July 30; the manuscript header and software release use July 29. DataCite confirms the arXiv DOI is registered and findable. The pinned author artifact is Apache-2.0 and includes source, prompts, paper configurations and compact evaluation evidence; ARC environment terms are separate. Python >=3.12, model/ARC access and a Docker or Finch sandbox are documented requirements.

## Primary sources

1. <https://arxiv.org/abs/2607.28287>
2. <https://arxiv.org/html/2607.28287v1>
3. <https://arxiv.org/pdf/2607.28287v1>
4. <https://api.datacite.org/dois/10.48550/arXiv.2607.28287>
5. <https://github.com/NIMI-research/Tycho>
6. <https://raw.githubusercontent.com/NIMI-research/Tycho/f68912a764372ead0a610db2e1c011d41ce5197e/README.md>
7. <https://raw.githubusercontent.com/NIMI-research/Tycho/f68912a764372ead0a610db2e1c011d41ce5197e/LICENSE>
8. <https://raw.githubusercontent.com/NIMI-research/Tycho/f68912a764372ead0a610db2e1c011d41ce5197e/pyproject.toml>
9. <https://raw.githubusercontent.com/NIMI-research/Tycho/f68912a764372ead0a610db2e1c011d41ce5197e/CITATION.cff>
10. <https://raw.githubusercontent.com/NIMI-research/Tycho/f68912a764372ead0a610db2e1c011d41ce5197e/PUBLIC_RELEASE_MANIFEST.json>

## About this entry

Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.
Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.