empirical-researchlisted

# Empirical Research Fires when **there is something here that could be false**. Documentation describes what a system does and there is no fact of the matter to be mistaken about. An empirical claim asserts the world behaves a certain way, so it can be wrong, overstated, contested, measured under conditions that are not the user's, or funded by whoever benefits. ## Tier **Light** (default): steps 1-6. **Deep** (the answer moves money, health, a public claim, or an irreversible decision): steps 1-6 plus every **Deep** addition marked below. The refutation pass is mandatory in **both** tiers. It is one query and it is the highest-yield step here. A run that skips it is web search wearing a lab coat. ## Protocol **1. Decompose.** Split the question into atomic claims that could individually be false. Not "should I take creatine" but "creatine increases strength in trained adults", "creatine causes kidney harm", "loading phases outperform steady dosing". Vague questions cannot be refuted, so they cannot be researched. **2. Retrieve primary sources.** See [Retrieval](#retrieval). Web search **orients only**. It never sources a claim. Pull the actual abstract before asserting what a paper found. **3. Tier every source** before leaning on it. See [Source tiers](#source-tiers). **4. Refutation pass (mandatory).** Run at least one query built to *break* your current conclusion, not support it. If you believe X works, search for X failing, X null results, X criticism, X overs