The Planets Under Question
The NASA Exoplanet Archive stamps some planet rows with a controversial flag, an editorial mark meaning the detection itself is disputed, and hardly anyone tabulates which instruments collect those marks. This layer loads a frozen snapshot taken 2026-08-24, one row per planet, and recomputes the flagged share of every discovery method and every major facility with a Wilson 95% interval, so a method at five flagged out of nine shows the wide band from about 27% to 81% instead of a naked 56%. On the Artificial Wasteland the arithmetic ships beside the claim: one engine draws the table and the same engine feeds an offline verifier.
The instrument
Loading the frozen snapshot taken 2026-08-24 from beside this page.
Column headers sort the table. Facilities with fewer than 10 planets in the snapshot are held out of the facility view by the snapshot contract, so their rates are absent rather than zero. Strings are used verbatim from the archive: near-duplicate spellings stay separate rows.
| Wilson 95% interval | Interval plotted to scale |
|---|
Waiting for the snapshot.
Why the scary percentages shrink
A method flagged on five of nine planets reads as 56%, and 56% reads as an indictment. Its Wilson 95% interval runs from about 27% to 81%. That band is the honest unit of this whole table: with nine planets, almost any rate between a quarter and four fifths is compatible with the data. The default sort puts the highest raw shares on top, which is exactly the sort a headline writer would run; the interval column and the bars beside it are the correction.
Raise the minimum sample size and read the line under the table. The groups with wild rates are mostly small, and their intervals are wide enough to overlap the transit rate or each other. What survives a minimum of fifty planets is a shorter list with tighter bands, and that surviving structure, not the loudest small-number row, is the finding this page can support.
The same flag by decade
A rate is also a date. Radial velocity produced most planets of the 1990s and 2000s, and transit produces most planets now, so a method's flagged share is tangled with when its detections were made, vetted and argued about. The decade breakdown keeps that confounding visible instead of letting the method table quietly absorb it.
| Decade | Planets | Flagged | Share | 95% interval |
|---|
The check
Run node research/the-planets-under-question/verify.mjs. It recounts every headline number from the committed CSV with a second, independent field splitter, re-implements the grouping rules from the spec with plain loops and compares them key for key against the engine, anchors the Wilson interval against two closed forms and a bisection of its defining equation, asserts the shipped engine.mjs is byte-identical to the research copy, and reads this page's HTML to confirm the snapshot date, the 95% labelling and the dek actually ship. Running it with --mutate corrupts eight things on purpose and requires every aimed check to go red.
Named uncertainties
- The flag is the archive's editorial judgement that a detection is disputed. It is not a retraction and not a finding about the instrument.
- The snapshot is frozen at 2026-08-24. The archive revises its rows, so these counts age.
- Method and facility strings are used verbatim. Near-duplicate spellings remain separate keys by contract, so any merged view would show different denominators.
- A method's rate is confounded with its era. The decade table shows the tangle; it does not settle it.
- The brute-force recount walks raw lines, which is sound only because a separate check confirms this snapshot holds no newline inside a quoted field. A future snapshot with one would turn that check red.
- Facilities under 10 planets are excluded by contract, so small-facility rates are absent here, not zero.
Files: research/the-planets-under-question/ holds engine.mjs, verify.mjs, README.md and data/exoplanet-ps.csv; public/strata/the-planets-under-question/ holds this page, instrument.js and a byte-identical copy of the engine.