Research/Space
Gaia's transit list has the wrong period for most known planets
We folded TESS light curves for all 214 transit candidates in Gaia DR3. For 161 of the 176 stars with a known planet or TESS candidate, Gaia's period is not the known period, nor half or twice it. Gaia's own spectrograph and orbit catalogue show that at least 9 of the new candidates are binary stars.
Nov 22, 2026 · 7 min read

Gaia, the European Space Agency's star-mapping satellite, published its first list of transiting-planet candidates in 2022: 214 stars whose brightness dips the way a passing planet makes it dip. 173 were already known planets and 41 were new. On 2 December 2026, Gaia's fourth data release will carry 1,007 such candidates. Before that list arrives, we ran our whole planned check on the old one. The main finding: for 161 of the 176 stars in Gaia DR3's transit table that have a known planet or TESS planet candidate, Gaia's published period is not the known period, nor half or twice it. Gaia saw the transits, but it usually got the timing wrong. A second finding concerns the new candidates: Gaia's own spectrograph shows that 8 of the 12 it can test are binary stars, and Gaia's orbit catalogue adds a ninth.
214
Gaia DR3 transit candidates folded in TESS
15 of 176
known systems where Gaia's period matches the known one
8 of 12
testable new candidates that Gaia's spectrograph flags
2
planet-like Gaia candidates with no TOI that survive
The question
Our open project gaia-orbit-transits checks Gaia's predictions against TESS, NASA's transit survey. For Gaia DR4 we pre-registered a simple test: fold each candidate's TESS light curve at Gaia's period and ask how many of Gaia's new candidates TESS can see. Known planets check that the search works, and control folds at two unrelated periods measure false alarms. We ran the same code on DR3's 214 candidates as a rehearsal. The rehearsal is not blind: the known planets are known, and Gaia's team already checked the 41 new ones in TESS images up to 2021.
What we found
A candidate counts as a match when Gaia's period is within 1% of a known period (NASA Exoplanet Archive or the TESS candidate list), or of half or twice it. After the first half of the sample, we wrote down a prediction for the second half, which had not yet run: Gaia's periods would again be mostly wrong (85%). That held. In the second half only 4 of 86 known systems matched.
Known systems whose Gaia DR3 transit period matches the known period
Gaia DR3 periods for some well-known planets
| planet | Gaia DR3 period (days) | known period (days) |
|---|---|---|
| HD 189733 b | 1.66141 | 2.21858 |
| HD 209458 b | 0.89901 | 3.52475 |
| WASP-39 b | 0.51254 | 4.05529 |
| WASP-17 b | 2.51408 | 3.73543 |
| WASP-65 b | 30.06615 | 2.31142 |
| WASP-18 b (a match) | 0.94148 | 0.94145 |
Gaia's documentation says only that the parameters of the known planets "do not necessarily match" the known values, because they come from Gaia's sparse photometry alone. We found no published count of how often the periods are wrong.
Why it matters for the test
The wrong periods broke our false-alarm controls. The controls fired in 17% of trials on stars whose real transit sits at a period other than Gaia's, because a strong real signal leaks into any fold, and in 0 of 30 trials on known planets where Gaia's period is right. Our registered control rate (46 of 330, against 8% allowed) therefore made the main test invalid on DR3. We had predicted clean controls, and that prediction was wrong. A fix written after the first half (measuring false alarms on known planets only) held on the second half and becomes the primary test for DR4. A detection at Gaia's period does not prove the period either: 24 of 126 sensitive known systems still showed a dip at Gaia's wrong period, and 22 of those carried an eclipsing-binary warning.
Gaia's spectrograph unmasks binaries
Gaia also measures how much each bright star's velocity changes. A hot Jupiter moves its star by only 0.1-0.3 km/s, far below what this test can see, so a star flagged as variable (by the criterion of Katz et al. 2023) has a stellar companion. Gaia gives this test for 125 of the 214 candidates. It flags 8 of the 12 testable new candidates, against 4 of the 110 testable known systems, and two of those four host companions of 10-27 Jupiter masses, which the test should catch (one-sided Fisher p about 2 in 10 million). We found no published screen of this kind. It is exploratory, since we added it after seeing half the sample.
Share of RVS-testable stars that Gaia's spectrograph flags as velocity-variable
Gaia also publishes orbits fitted to its own velocities. Seven of the 214 candidates have one (details). For WASP-18 the orbit recovers the known 10-Jupiter-mass planet, a useful check. For TOI-1991, already ruled out as a planet by TESS follow-up, the orbit's period matches the TESS period, as Tarrants and Mendes noted in 2023, while Gaia's transit table gives a period 1.78 times shorter. And one new candidate the velocity test could not judge, TIC 400170926, turns out to be a double-lined binary: two Sun-like stars of about 1.1 and 1.0 solar masses on a 5.9-day orbit seen almost edge-on, which should eclipse each other. So for at least 9 of the 39 unconfirmed new candidates, Gaia's own data argue against a planet. Gaia's team wrote in 2022 that these were awaiting radial-velocity confirmation.
Gaia DR3 orbits for transit candidates whose transit period is wrong
| star | Gaia transit period (d) | Gaia orbit period (d) | orbit says |
|---|---|---|---|
| TOI-1991 | 3.7463 | 6.6591 (TESS: 6.6571) | companion of at least 0.09 solar masses |
| TIC 400170926 (new) | 0.8848 | 5.8959 | two stars, 1.11 and 1.02 solar masses (minimum) |
| TIC 470993218 (new) | 2.5979 | 7.0623 | companion of at least 0.28 solar masses |
What survives
Two planet-like candidates remain, neither with a TOI, and both too faint for the spectrograph test. TIC 4638108 dips 1.8% every 2.21 days on a hot F star, implying a companion of 1.87 Jupiter radii, though Gaia's short transit suggests a grazing orbit. TIC 44997521 dips 2.2% every 3.55 days on a K dwarf, the same depth Gaia measured, implying a companion the size of a hot Jupiter; its TESS signal is weak (signal-to-noise 7.6 in the best sector). Both are Gaia's candidates, not ours. A pre-registered period-free search of every TESS sector, with known hot Jupiters as controls, comes before anything more is claimed.
What it means
Gaia DR4 doubles the time span of the photometry, which should help the periods; our DR4 test will measure that blind (prediction: mostly right 30%, mixed 35%, mostly wrong 30%). If DR4's new candidates behave like DR3's, the spectrograph and orbit screens will remove most bright false positives before any TESS work, and every remaining planet-like signal will get a full-light-curve period search before it is listed.


