Fable is gone but the loop isn't (github.com)

Grainulation has introduced "bean," a new recursive convergence loop designed to enhance task execution within AI frameworks like Claude Code and Codex. Unlike the now-defunct Fable model, which mandated a fixed sequence of phases, bean focuses on continuous investigation and revision. It evaluates ongoing tasks by examining unresolved conflicts and weak evidence, looping through these stages until a decisive answer is reached. This approach encourages a dynamic exploration of data, adapting to identified gaps without pre-defined limits, ultimately allowing AI systems to self-correct and build more robust conclusions. The significance of bean lies in its potential to transform how AI models manage complex tasks. By leveraging a claim ledger for memory and a compiler to score convergence, bean ensures that the AI iteratively refines its outputs. It enables independent checks for high-stakes tasks, allowing for enhanced reliability and rigor in evidence-based decision-making. With no dependencies or network requirements, bean simplifies integration and maintains privacy, appealing to developers seeking efficient, autonomous problem-solving tools within the AI/ML landscape. This innovation marks a substantial step forward in creating models that learn and adapt in real-time, setting a new standard for task orchestration in AI environments.