research-brieflisted

# Research Brief A question in, a decision-shaped research plan out. Methods exist to move decisions — any method that doesn't map to one gets cut. ## Verification gates (defined first; output is blocked until all pass) - **G1 — No orphan methods:** every proposed method maps to a named decision and states which answer would push the decision which way. A method with no mapping fails the gate. - **G2 — No starved decisions:** every decision option has at least one method capable of ruling it in or out. An option no method can touch fails the gate. - **G3 — No fabricated context:** the plan assumes only access, data, and users stated in the input (or explicitly flags each extra assumption as a prerequisite to confirm). ## Steps 1. **Extract the decision.** What will be done differently depending on the answer? If the input has a question but no decision, ask for the decision — research without a decision is a reading list. Write the decision options and the criteria ("we roll back if …"). 2. **Decompose the question** into the 2–4 sub-questions the decision actually turns on (what changed, for whom, why, would fixing it work). 3. **Pick methods per sub-question,** cheapest-decisive first: existing analytics → targeted interviews/replays → surveys → experiments. For each: participants/data source, sample size with its basis (n=5 because pattern saturation; n=200 because ±7% at 95% — never a naked n), time/cost class, and the kill/confirm signal. 4. **Build the method→decis