meta-reasoning

FOL Software Archaeology

Contains code for expansion policies, decision-making and backup rules, and h-cost learning strategies for combining to make real-time heuristic search algorithms.

Meta-Problem Modeling (MPM) cognitive flywheel as a resident engine for LLM agent runtimes. 六阶段问题生命周期 · ε-δ 双判据收敛 · 代谢账本 · 导师规则层

Hybrid AI routing: LOCAL Ollama + CLOUD GitHub Copilot

Reasoning suite for coding agents: a scored meta-router plus nine reasoning-pattern skills (ToT, BoT, SRC, HE, AR, DR, AT, RTR, NDF) with parallel-execution and handover infrastructure — gauntlet-validated backstage discipline

Agent C: cloned from the v1.0 DEV baseline. Future specialization track for synthesis, hypothesis generation, tool orchestration, and long-form reasoning workflows in the cluster.

Logs and evidence for the ANM-V1.5 prototype: Web-of-Thought engine, parallel R1 workers, verifier output, and system traces executed on a MacBook Air M2.

Canonicals for terminating unproductive reasoning loops in language models, including recursive self-reference, consciousness spirals, and language–stakes conflation.

The PLM harness: the model authors its computation as policies. REPL kernel, llm/react_auto/react primitives, typed constraints, message and state weaving.

This system models adaptive reflection across multiple cognitive layers, enabling agents to restructure their reasoning through a meta-level feedback process. It enhances self-regulation and decision robustness in dynamic environments. 本システムは、複数の認知層にわたる適応的反射をモデル化し、エージェントがメタレベルのフィードバックを通じて推論を再構築できるようにします。これにより、動的環境下での自己調整力と意思決定の堅牢性が向上します。

A self-referential knowledge system combining GEB, Chomsky, and Leibniz for structural AI