← ClaudeAtlas

factfulnesslisted

A data-thinking skill based on "Factfulness" (Hans Rosling et al.). Audit every statistic against four provenance items (source / year / definition / baseline), diagnose the ten dramatic instincts, and output a level-plus-trend dual view and a danger-by-exposure fact-check card. Use when the user says "is the world getting worse", "is this number trustworthy", "cite a statistic for me", "stop fear-mongering", "let the data speak", "do we have to decide right now"; also for any output involving data citation, trend judgment, risk communication, or urgency-driven decisions.
Trustbaselab/trustbase-skills · ★ 1 · AI & Automation · score 74
Install: claude install-skill Trustbaselab/trustbase-skills
# Factfulness · An Operating System for Citing Data **The core claim of *Factfulness*:** our judgments about the world are systematically worse than random. In a 2017 study across 14 countries, the average score on 12 global-fact questions was **2.2 out of 12**, while random guessing would score **4 out of 12**, and 15% of respondents got every question wrong. That cannot be ignorance — it is a **systematically overdramatic worldview**, shaped by **ten dramatic instincts plus the media's appetite for bad news**. **This skill translates the book's central rule — "look at the world through data, not through emotion and bias" — into concrete data-acquisition and citation discipline for an agent.** ## When to use - The user asks "is the world getting worse", "is X finished / collapsing" — a level-plus-trend audit is needed - The user asks you to cite or verify a statistic, a trend, or a market judgment - The output touches risk, disaster, or panic themes (including self-restraint against fear-mongering) - The context contains urgency cues such as "must right now / immediately / or it's too late" - The user frames a binary judgment ("poor vs rich", "developed vs undeveloped") - An attribution discussion ("it's all because of X", "we owe everything to Y") - The agent itself is about to cite data or cases from training memory **Lightweight mode**: when only one or two numbers are being verified, or the only question is "is this number trustworthy", output **a one-sentence concl