ANCHOR: Branch-Point Data Generation for GUI Agents

ANCHOR addresses the primary bottleneck in desktop GUI agent development: the scarcity of high-fidelity interaction data. While web-based agents benefit from HTML structure, desktop agents must rely on visual pixels and complex OS interactions, making human demonstrations expensive. ANCHOR's methodology of using 'branch points' to bootstrap data is intellectually sound and represents a high-value technique for the current agentic research cycle. However, the defensibility is low (score 3) because it is primarily a research methodology; once the paper is public, the 'branch-point' logic can be replicated by any well-funded lab. The frontier risk is 'high' because OpenAI (Operator), Anthropic (Computer Use), and Google (Jarvis) are all aggressively pursuing desktop automation. These labs likely already use similar self-play or trajectory expansion loops internally. The 0-star count reflects its extreme recency (5 days), while the 4 forks suggest immediate peer interest in the research community. For an investor, the value is not in the software itself, but in the potential for this technique to be the engine behind a larger open-source dataset project (like a 'ShareGPT for Desktop').