False flagging “reasoning_extraction” is real problem

A user reports persistent issues with Anthropic's Sonnet 5.5 Max and Opus 5.5 Extra models, which are blocking analytical tasks due to "reasoning_extraction" flags. Despite checking rules and ensuring prompts contain no triggers, the user's work, including analyzing technical documents and drafting feedback, is repeatedly halted. This has wasted significant time and credits, forcing a switch to GPT-6 Pro. The user expresses frustration, noting that while systematic extraction blocking is understood, the current implementation is flawed and unacceptable.