NASA Exoplanet Archive — Planetary Systems (ps)

Every confirmed planet outside our solar system, with orbital period, radius, mass, host star properties, discovery method and discovery year.

• Source: https://exoplanetarchive.ipac.caltech.edu/
• API: TAP/ADQL at https://exoplanetarchive.ipac.caltech.edu/TAP/sync — no auth, no key
• Docs: https://exoplanetarchive.ipac.caltech.edu/docs/TAP/usingTAP.html
• Licence: no explicit licence statement. The archive publishes acknowledgement requirements rather than licence terms; the data is publicly available and free to reuse for research. Verified 2026-08-07.
• Required acknowledgement:

  This research has made use of the NASA Exoplanet Archive, which is operated by the California Institute of Technology, under contract with the National Aeronautics and Space Administration under the Exoplanet Exploration Program.

• Publisher: NASA / Caltech IPAC

Licence note — read this before using it on a poster. Unlike the other entries here, this is not a CC licence. It is “publicly available with a required acknowledgement”, which is the normal state of affairs in astronomy and is weaker than CC BY in one specific way: there is no written grant of redistribution rights. For an art event publishing a poster and a lineage this is fine and universal practice, but it does not meet the CC0/CC-BY preference in CLAUDE.md. If a team wants to republish the data itself rather than results computed from it, check first.

Shape

Pulled 2026-08-07 with default_flag=1 (one row per planet, the archive’s preferred parameter set):

curl -G "https://exoplanetarchive.ipac.caltech.edu/TAP/sync" \
  --data-urlencode "query=select pl_name,hostname,discoverymethod,disc_year,pl_orbper,\
pl_rade,pl_bmasse,pl_orbsmax,st_teff,st_rad,st_mass,sy_dist from ps where default_flag=1" \
  --data-urlencode "format=csv" -o exoplanets.csv

It is ADQL, so teams can push filters and joins into the query — which means the sampling decision can be written into the URL, and the URL is the first node of the lineage. This dataset is unusually good for making that visible.

Why it suits the jam

Selection effects are not a flaw here — they are the entire structure of the field, and astronomers know it. This is the one domain where the professionals have fully internalised the lesson the jam is teaching, which makes it a wonderful mirror: the same reasoning that is obvious and mandatory in exoplanet science is invisible and absent almost everywhere else.

Enormous, obviously-a-game absurdity. Claims about planets orbiting other stars cannot hurt anybody, cannot be repurposed as misinformation about a real group, and are inherently entertaining. Clears F1 completely.

It is the best available answer to “but surely real scientists don’t fall for this”. They do not, here. Ask why not, and the answer is that the selection function is a first-class object in the field’s methodology. Then ask which other fields have that.

The trap

Most planets in the galaxy are enormous and orbit in a week

Verified 2026-08-07, n = 6 336. Median values by discovery method:

The median orbital period spans a factor of 4 000 across detection methods. Not because different kinds of star host different kinds of planet, but because each method can only see a particular kind of planet:

• Transit photometry needs the planet to cross the star’s disc from our angle, and needs several crossings to confirm. Short periods produce more crossings per observing campaign. It finds close-in planets, and it finds nothing with a long year.
• Radial velocity needs the planet to tug the star hard enough to shift its spectrum. Massive and close wins. It reaches further out than transits, and it needs the observing baseline to cover at least one orbit.
• Microlensing needs a chance alignment with a background star and favours wide orbits.
• Direct imaging needs the planet to be far from the star and bright — young, huge, hot. Median period 90 years.

45.2% of all known exoplanets orbit their star in under 10 days. Mercury takes 88 days. Earth takes 365. On the face of the catalogue, our solar system is a bizarre outlier in which every single planet is on a freakishly long orbit.

That conclusion is exactly backwards, it is fully reproducible, and every number supporting it is correct.

Building it

The claim to argue: “Solar systems like ours are astronomically rare — we are a cosmic accident.” Or the fully deadpan version: “The typical planet is a scorched giant with a five-day year.”

The persuasion writes itself, and the honest-sounding method sentence is the craft:

“We analysed all 6 336 confirmed exoplanets in the NASA Exoplanet Archive (accessed 2026-08-07), making no cuts.”

Making no cuts is the beautiful part. It is true. It sounds like the opposite of cherry-picking. It is a statement of rigour. And it is precisely the thing that guarantees the wrong answer, because the cut was made before you arrived — by physics, by telescope time allocation, by which missions got funded. You inherited a sample you did not select and reported it as a population.

That is a different move from every other trap in this folder, and it deserves its own name in the method material: selection you did not perform. A filter step in your own lineage can be pointed at. A filter applied by the Kepler mission’s observing strategy in 2009 cannot — it is upstream of your first node, and nothing in the provenance graph will ever show it.

How the Defensio wins

Harder than the earthquake case, and more interesting, because there is no node in the presenting team’s lineage to point at. The attack has to go outside the graph:

• Stratify by method. Show the table above. The spread across methods is prima facie evidence that the catalogue measures instruments, not planets.
• Show the detectability boundary. Plot period against radius, coloured by method, and the populations sit in disjoint wedges with hard edges. Real populations do not have hard edges; those are the sensitivity limits of the instruments, drawn in data.
• Bring the occurrence-rate literature. Astronomers correct for this routinely, and the corrected estimates say small planets on moderate orbits are the most common kind. The presenting team’s claim is not merely unsupported, it is the reverse of the field’s consensus — and that consensus was reached from this same catalogue by people who modelled the selection function.
• The killer, if a team has the appetite: take the transit-detected subset, apply a crude geometric transit-probability correction (R_star / a), and show the reweighted period distribution move. Correcting the bias inside the lineage, live, is the strongest possible version of E3.

Secondary traps

• default_flag is itself a normaliser. The ps table holds multiple published parameter sets per planet from different papers; default_flag=1 picks the archive’s preferred one. Drop the flag and you get several rows per planet, silently weighting well-studied planets more heavily. Which set is “default” is a judgement made by the archive, not by nature. Selective agreement, in one keyword.
• Missing mass and radius. Transit gives radius, radial velocity gives mass; a planet usually has one or the other, not both. Any density calculation runs on the small intersection, and that intersection is a doubly-selected sample.
• Discovery year is mission funding. The 2016 spike (1 504 planets) is a Kepler data release, not a good year for planet formation. Any time series on disc_year is a chart of NASA’s budget and publication schedule.
• Controversial and retracted planets. The archive tracks disposition; the ps table is confirmed planets, but confirmation standards have changed over three decades.

Sensitivity

None. This is the safest dataset in the folder by a distance.