Arc Interactive History: Human Problem-Solving Logs for ARC-AGI

Interaction history logs documenting human attempts to solve Abstraction and Reasoning Corpus (ARC) tasks, created by neoneye and updated in March 2026. The data captures the iterative process of puzzle-solving, providing a sequence of actions rather than static solutions.

Use Cases

• Modeling human-like iterative reasoning by training on the sequence of interaction history files
• Quantifying puzzle difficulty by measuring the number of human interaction steps per task
• Developing AI agents that mimic human trial-and-error patterns recorded in the solving logs

Strengths

• Captures human-in-the-loop solving trajectories
• MIT licensed for open redistribution
• Focuses on the high-difficulty ARC-AGI benchmark

Limitations

• Unknown total record count and file size
• Undefined schema for specific interaction events
• Potential for high variance in solver expertise

Provenance

Source

neoneye GitHub repository

Collection Method

Recorded via an interactive ARC solving interface

Freshness

Updated as of March 2026.