researchlisted

# AEP — Research Protocol **Hard rule: memory is a hypothesis, the web is evidence, and one source is an anecdote.** Anything version-specific, fast-moving, or decision-critical gets researched, not recalled. ## 1. Decompose before searching State the research question in one line, then split it into its **sub-branches** — the questions that must each be answered for the main answer to stand (mechanism, alternatives, limits, security posture, maturity/maintenance, version compatibility with §P.1). A topic is not researched until its load-bearing sub-branches are. Gate yourself Self-Ask style: *"What must be true for my current belief to hold — and which of those have I actually checked?"* Unchecked items are the search queries. ## 2. Search from multiple angles One phrasing finds one community's answer. For each sub-branch, vary the angle: official docs/changelog · the source itself (repository code, issues, release notes) · academic (arXiv/ACL/conference) where the claim is scientific · practitioner reports (postmortems, benchmarks, discussions) · **the negative angle** ("X problems", "X vs Y", "migrating away from X"). If every result agrees suspiciously, you searched one angle. ## 3. Triangulate or label - A claim used for a decision requires **≥2 independent sources** — independent means not citing each other and not the same author/vendor. Vendor claims about the vendor's product count once, at half weight. - Separate **claim / evidence / opinion** while reading.