mode-routinglisted

# Open or clamped Two failure modes, opposite causes, and the setup has to be able to produce both. > "If you are too specific, Claude will follow your instructions even when a pivot may be > more appropriate. If you are too vague, Claude will often make choices and assumptions > based on industry best practices that may not be a fit for your task." > — Thariq Shihipar, *Finding Your Unknowns* > "Examples fence Claude into the demonstrated space." > — Thariq Shihipar, *The new rules of context engineering for Claude 5* A wall of rules does not make work safer in general. It makes it **conventional**. That is the right trade when the output is a published number and the wrong one when the output is an idea you do not yet have. ## Which mode | The output is | Mode | Why | |---|---|---| | A research direction, an experiment design, a hypothesis, a name, an API surface, a diagnosis with no known root cause | **OPEN** | The answer is not in the repo. Constraints here cost you the answer. | | A number that will be published, a rebuttal sentence, harness or spec code, anything under the artifact freeze, anything a reviewer reads | **CLAMPED** | The answer exists and must be found exactly. Invention here is fabrication. | | Refactors, config, tooling, docs | **CLAMPED, cheaply** | Known-good shapes exist. Follow them and move on. | The mistake to avoid is treating *research* as clamped because it is important. Importance selects for rigour in the **checking**, not for narrowne