researchlisted

# First-Principles Research Engine This skill exists because AI research defaults to a specific failure mode: producing text that *looks* rigorous — good formatting, citations, logical flow — but rests on authority bias, lazy analogies, and narrative coherence rather than mechanism analysis. The skill prevents this through structural gates: mandatory steps that must be completed before proceeding. Structure beats instruction. "Think more rigorously" changes nothing. "State the prior before researching, generate three competing hypotheses before evaluating any, actively search for disconfirming evidence before concluding" produces measurable improvement. ## Core Design Principles These five principles (derived from analysis of STORM, GPT Researcher, ReAct, claim decomposition systems, and gap-reflection architectures) shape every phase of the protocol: 1. **Separate question generation from answer generation.** The quality of research is bounded by the quality of questions asked. Generating questions and answers in the same pass means questions are shaped by what you already know — circular research. The question-engineering phase must complete before investigation begins. 2. **Enforce perspective diversity before convergence.** A single perspective produces a single frame. Multiple perspectives produce genuine coverage. The perspectives must be structurally different (economics vs. psychology vs. technology), not superficially different (slightly reworded versions of the