Verifiable Thinking

Acerca de

Addresses the inherent weaknesses of Large Language Models (LLMs) in complex reasoning by providing a robust MCP server. This tool proactively detects common cognitive traps (e.g., bat-and-ball problem, Monty Hall) with O(n) heuristics, preventing predictable LLM failures before they occur. It offers a structured scratchpad for multi-step reasoning, complete with confidence tracking, auto-verification gates, and local mathematical computation. Furthermore, it incorporates query-aware context compression to manage long reasoning chains efficiently, ensuring more reliable, verifiable, and cost-effective AI output.

Características Principales

• Local math expression evaluation to avoid unnecessary LLM round-trips
• Structured reasoning scratchpad with step tracking, confidence scoring, and automatic verification gates
• Query-aware context compression (CPC-style) for efficient handling of extensive reasoning chains
• 15 cognitive trap pattern detectors (e.g., additive system, Monty Hall, base rate) with sub-millisecond latency
• Algebraic error detection and progressive simplification hints
• 0 GitHub stars

Casos de Uso

• Improving the reliability and accuracy of LLMs in solving complex problems and analytical tasks.
• Preventing common cognitive biases and logical fallacies in AI-generated responses.
• Structuring and verifying multi-step reasoning processes for AI agents and applications.
• Reducing token usage and operational costs for LLM-powered systems by offloading verification and context management.