Eleven datasets, all real and all publicly licensed. Each page records where it came from, what licence it carries, how to fetch it, and — the part that matters here — which trap it sets.
Pick one that is boring enough to be trusted. Nobody suspects a thermometer, and an audience that would push back instantly on crime or health figures will accept a weather chart while it is used to argue something ridiculous. The absurdity should come from your argument, never from the data being silly.
Two warnings worth reading before you choose. Check the licence — it is on every page, and several of these carry required attribution text you must reproduce. And prefer subjects where an absurd conclusion is obviously a game: material that misleads about real people or real health outcomes can escape the room.
They are already here
You do not have to fetch anything. Every one of these is in data/ in the repository, so the whole
team starts from the same bytes and a hash means the same thing for everybody.
| file | dataset |
|---|---|
murder.csv | homicide and inequality, 46 countries |
fog-nebel-gew.csv | GeoSphere, fog and thunderstorm days, 10 stations, 1990–2025 |
quakes.csv | USGS, California, M3+, 1970–2025 |
penguins.csv | Palmer Penguins |
indo-rct.rda | Indomethacin RCT — load() it; no CSV of it is published |
nasa-exoplanets.csv | NASA Exoplanet Archive, confirmed planets |
owid-life-expectancy.csv + .metadata.json | Our World in Data — keep the metadata, it carries the licence |
pharmaverse-adsl.rda | pharmaverse ADaM, subject level |
wien-foerderbericht-2024.csv | Vienna subsidy report |
wien-bezirksgrenzen.json | Vienna district boundaries |
wien-baumkataster.csv.gz | all 232,770 Vienna trees — gzipped; read.csv() opens it directly |
Three of them arrived as a slice of something larger, and the slice is a sampling decision, so
it is written into bin/fetch-data beside the fetch rather than left to memory. The tree register
is gzipped rather than cut down, because 53 MB is a lot to clone on event wifi and choosing a
subset would be choosing a sample on your behalf.
CORDIS is the exception — the EU project archives are multi-hundred-megabyte zips. Take the slice you want from cordis.europa.eu/about/archives and record that pull as its own step.