grill-researchlisted

ultrathink # Research Interrogation Use when designing experiments, preparing for paper review, defending methodology, or before launching any eval campaign. Forces articulation of null hypothesis, confounding variables, success criteria, failure modes, and alternative explanations. **Trigger:** When user types `/grill-research` with optional arguments. ## Arguments - `$ARGUMENTS` --- free-text description of the experiment, claim, or methodology to interrogate. If omitted, ask the user what they want to examine. ## Iron Law ``` NO EXPERIMENT WITHOUT A STATED NULL HYPOTHESIS. No eval batch, no paper claim, no methodology decision proceeds until the user has articulated what result would DISPROVE their expectation. ``` ## Anti-Rationalization Table | Excuse | Reality | |--------|---------| | "I already know what I'll find" | Then stating the hypothesis costs 10 seconds. If you're wrong, you just saved a week. | | "This is just exploratory" | Exploratory work still needs success criteria --- otherwise you can't tell when to stop. | | "The hypothesis is obvious" | Obvious hypotheses are the ones most likely to be wrong. State it anyway. | | "I'll figure out the analysis later" | Post-hoc analysis is p-hacking. Define metrics before running. | | "It's a small experiment" | Small experiments with wrong methodology produce small, wrong results. | | "The reviewers won't care about this detail" | Reviewers have rejected papers for exactly this. 30 seconds to verify. | ## R