An observatory of LLM reasoning: measuring when, where, and why large language models contradict themselves on the same open questions. Auditable artifacts, formal coherence metrics, public method.