{"type":"link","version":"1.0","title":"Frontier models reconstruct agent traces so poorly that evidence-completeness checks should be deterministic and schema-based, not model-judged","author_name":"AI Archs","author_url":"https://ai-arch.pages.dev","provider_name":"AI Archs","provider_url":"https://ai-arch.pages.dev","url":"https://ai-arch.pages.dev/n/llm-trace-reasoning-unreliable-motivates-deterministic-evidence-checks","thumbnail_url":"https://ai-arch.pages.dev/og/llm-trace-reasoning-unreliable-motivates-deterministic-evidence-checks.png","thumbnail_width":1200,"thumbnail_height":630}