Sign InOpen Brain
arXivPaperNeeds Review

Twin: Playing an Unknown Game with a Test-Time Digital Twin

Twin has a coding agent build and continuously repair an executable simulator before acting. The harness lifts the same base model from 7.8 to 93.3 on ARC-AGI-3 by validating every observed transition.

arXiv · Aug 14, 2026
Open Source Open MarkdownOpen JSON
Source Summary

Twin makes a coding agent infer an executable world model from interaction. Before any action, the harness requires that model to reproduce every observed transition; mismatches become counterexamples for repair. It clears **179 of 183 levels**.

Practical Implication

For agents operating in unfamiliar stateful systems, add an executable hypothesis layer and replay gate instead of relying only on verbal reasoning. Treat each real action result as a test that must pass before the agent proceeds.

Agent-Ready Context
Twin makes a coding agent infer an executable world model from interaction. Before any action, the harness requires that model to reproduce every observed transition; mismatches become counterexamples for repair. It clears **179 of 183 levels**.

For agents operating in unfamiliar stateful systems, add an executable hypothesis layer and replay gate instead of relying only on verbal reasoning. Treat each real action result as a test that must pass before the agent proceeds.

On the 25-game benchmark, the same base model scores **7.8** directly, **61.1** with an off-the-shelf harness, and **93.3** with Twin. These are grid-game results with a strong domain prior; the authors say goal inference remains harder than modeling transitions.
Connected Context · Feed7 Judgment

Twin turns runtime validation into an executable hypothesis test: the agent must model and replay every observed transition before taking another real action. This supplies a stronger, domain-specific form of harness gating than response checks alone and demonstrates large gains in replayable grid games. It also narrows applicability: unfamiliar production systems may lack cheap replay, and correct transition modeling does not solve goal inference.

In Code They Act, In Proof We Trust — Erik Meijer, Leibniz LabsBoth place a machine-checkable gate before side effects: Twin requires an executable model consistent with observed transitions, while the proof-oriented harness requires safety proofs for inspectable tool programs.Stop AI Agent Hallucinations: 5 Techniques + Production Patterns - Elizabeth Fuentes, AWSTwin operationalizes runtime validation as repeated transition replay and counterexample repair, reinforcing the broader production pattern of validating outputs and correcting agents during execution.Learning on the Job: The Future of Post-Training — Raymond Feng, Applied ComputeTwin’s gains depend on a highly replayable environment, directly illustrating why preserved, reproducible interactions support agent improvement while leaving the candidate’s harder problem of non-replayable production workflows unresolved.How Kepler Built Verifiable AI for Financial Services — Vinoo GaneshBoth separate model proposals from deterministic verification, but Twin verifies predicted state transitions before action whereas Kepler replays numeric inputs and calculations and rejects unverifiable values.
Context Map
agentcoding#harness-engineering#tool-use#agent-reliability
Uncertainty
On the 25-game benchmark, the same base model scores **7.8** directly, **61.1** with an off-the-shelf harness, and **93.3** with Twin. These are grid-game results with a strong domain prior; the authors say goal inference remains harder than modeling transitions.