Concept-Guided Chain-of-Thought (CGCoT) pairwise annotation tool for systematic text evaluation using LLMs. Generate breakdowns, compare items, compute scores, and validate against human judgments. Supports Ollama, Hugging Face, Google Gemini, OpenAI, and Anthropic models.
Bradley-Terry leaderboards with simultaneous confidence control — across 106,134 LMArena votes only 1 of 54 neighbouring pairs survives, against 5 when each is judged on its own.
