What Qwen is thinking when it gets stuck saying "Wait"

MIT CSAIL had GPT-6 Astra fix Qwen3.5-0.8B's reasoning loops by editing 0.016% of its weights (MATH-500 went from 7.9% to 48.1%). They released the diff, so the edit can be rebuilt exactly and run locally. I took it a step further and had Claude look at Qwen with Anthropic's Jacobian lens. At the token where the model decides to write "Wait" again, the lens reads "Wrong", "Nope" and "Hmm" from about 60% depth on, even though the answer is already written out. At real stopping points it reads "", "Answer" and "Output". With the edit applied, the lens at the same token shows "Verification", "Completion" and "Result" in place of the doubt. The edit converts to GGUF cleanly (the base and edited bf16 files differ only in the two edited tensors) and still holds at Q4_K_M, with 1 loop in 40. I ran the evals in llama.cpp (Vulkan) on a Ryzen AI 9 HX 370 mini PC. Samples are small, but here's the code: github.com/alisorcorp/inside-a-reasoning-loop/tree/main/code And data: alisorcorp.github.io/inside-a-reasoning-loop