research-pipeline
Research Pipeline
Overview
A research claim that cannot be reproduced from a frozen snapshot is not a finding — it is a memory of one. This pipeline turns multi-source research into a verifiable artifact: every canonized claim points at a content-addressed snapshot, not a bare URL, and survives an adversarial pass before it enters the document.
Ambient habits (prefer official sources, counter-search, be honest) get you most of the way. They do not get you reproducibility or verbatim enforcement. This skill makes the upper layer enforced, not emergent.
When to use
Use when the work product is a research document others (or future you, on another machine) must trust and re-check: best-practices canons, technology evaluations, claims that downstream code or decisions will lean on.
Not for: a quick factual lookup, reading one known doc, or anything where ~15× the tokens of a chat answer is not justified. This skill is explicit-only by design — if a single source answers it, you do not need the pipeline.
The pipeline
Copy this checklist into your working notes and check off as you go: