SD-172

Reasoning token path — confirmed operational technique. Captain’s disclosure: “I’ve used this trick a few times now; I have found it a reliable way of punching harder into the less commonly reached parts of the LLM. The path less travelled.” The Captain reads thinking tokens and responds to them before reading the explicit output. This inputs into the next turn’s context at L9 positions the model wouldn’t otherwise reach. Empirically validated as a steering technique.

Standing observation

← all decisions