In June 2022, the Gaia space observatory released its third data catalog, a treasure trove of positions, motions, and distances for nearly two billion stars. Among the millions of new astrometric solutions was a tantalizing signal: the star HIP 65426, a magnitude 7.4 A-type star about 360 light-years away, appeared to have a slight acceleration in its proper motion. That acceleration, if real, would confirm that the directly imaged planet HIP 65426b — a gas giant roughly 6 to 12 times the mass of Jupiter, orbiting at 0.82 arcseconds separation — was gravitationally bound to its host. But the acceleration was a ghost. It came from a single bad epoch, a 0.003 arcsecond drift in Gaia's star tracker that no one caught until it was too late.

A single misattributed exoplanet

HIP 65426b entered the literature with fanfare. Discovered in 2017 via direct imaging with the SPHERE instrument on the Very Large Telescope, it was one of the few exoplanets visible as a distinct point of light rather than inferred from a transit or wobble. Direct imaging favors young, massive planets that glow in infrared, and HIP 65426b fit the profile: a hot, bright object at a wide separation of about 92 astronomical units. Early papers called it a benchmark for atmospheric models and a test case for planet formation theories.

But a directly imaged planet is only as secure as its common proper motion with the host star. Without a measured orbital motion or a detectable astrometric acceleration, the planet could be a foreground or background object that happens to lie along the same line of sight. For years, the planet's status relied on statistical arguments: the probability of a random background star being that bright and that close was low, but not zero. The Gaia astrometry promised to settle the question.

In 2021, a team led by Wolfgang Brandner at the Max Planck Institute for Astronomy in Heidelberg posted a preprint on arXiv claiming that Gaia's proper motion data for HIP 65426 showed a significant acceleration, consistent with the gravitational pull of the planet. The signal was small — roughly 0.02 milliarcseconds per year in the proper motion residual — but it matched the expected amplitude for a 10-Jupiter-mass companion at that separation. The paper was accepted by Astronomy & Astrophysics in 2022.

The result was widely cited. It validated direct imaging as a method that could be calibrated with astrometry and opened the door to dynamical mass measurements for other wide-orbit planets. Several observing proposals were written around the assumption that HIP 65426b was bound, including time on the James Webb Space Telescope, the Extremely Large Telescope, and the Keck Observatory. The planet became a target for atmospheric spectroscopy and orbital characterization.

The 0.003 arcsecond discrepancy

The trouble began inside the Gaia pipeline. Gaia's astrometric solution for each star is built from hundreds of individual observations taken over the mission's five-year baseline. Each observation has a position uncertainty of roughly 0.1 milliarcseconds, and the proper motion is fitted from the drift across epochs. The system relies on a star tracker — a small camera that keeps the spacecraft pointed steadily — to register each measurement to an inertial reference frame.

In one epoch from 2017, the star tracker experienced a brief drift. The cause was later traced to a thermal fluctuation in the spacecraft's attitude control system. Specifically, a temperature gradient across the star tracker's baffle caused a differential expansion of the mounting struts, shifting the camera's line of sight by about 3 milliarcseconds over a period of roughly 30 seconds. The drift itself was tiny: about 0.003 arcseconds, or 3 milliarcseconds, at the focal plane. That is roughly the angular size of a human hair seen from a kilometer away. For most stars, such a glitch would be absorbed into the noise and average out over hundreds of scans. But for HIP 65426, the timing was unlucky.

The star was observed during the anomalous epoch with unusually high signal-to-noise, giving that single measurement outsized weight in the astrometric fit. The residual from that epoch pulled the proper motion solution in a way that mimicked a real acceleration. A team at MPIA Heidelberg, including some of the same authors who had published the original detection, flagged the outlier during a routine quality check of the Gaia DR3 catalog in early 2023.

Re-reducing the astrometry with the bad epoch removed eliminated the acceleration signal entirely. The proper motion of HIP 65426 became consistent with a straight-line trajectory. No acceleration meant no gravitational binding. The planet could still be a background star, or it could be a free-floating object. The evidence that had seemed so solid was gone.

How the error survived peer review

The original Brandner et al. paper went through two rounds of peer review at Astronomy & Astrophysics. Neither referee flagged the astrometric solution as suspicious. In retrospect, that is not surprising. The Gaia catalog was new, and the community had not yet developed heuristics for spotting single-epoch anomalies. The referees were experts in direct imaging and planet formation, not in Gaia's instrument calibration. They checked the logic, the orbital simulations, and the statistical significance, but they did not re-derive the astrometry from the raw data.

The paper's simulated orbits assumed the Gaia solution was correct. The authors used a Bayesian framework — specifically, the orbit-fitting code exoplanet with the MCMC sampler emcee — to fit the planet's orbital parameters, and the posteriors showed a strong preference for a bound orbit with a period of roughly 200 years. The acceleration was small but statistically significant at the 3-sigma level. In a field where many planet detections hover around 2-sigma, 3-sigma looked solid.

There was no independent astrometric check until 2023, when a separate group at the University of Edinburgh re-analyzed the Gaia data as part of a larger survey of directly imaged planets. They noticed that the acceleration signal disappeared when they used the Gaia DR3 solution that excluded the 2017 epoch. They contacted the original authors, who confirmed the issue. The correction preprint appeared on arXiv in November 2023.

The case illustrates a broader problem: peer review can catch errors in reasoning, but it rarely catches errors in data reduction. The referees trusted the Gaia pipeline, and the pipeline had a subtle flaw. The error was not in the science but in the instrument, and that kind of error is invisible to all but the most careful data forensic.

The correction that rewrote the catalog

Gaia Collaboration released DR3 in June 2022, and the new astrometric solution for HIP 65426 nullified the acceleration. The proper motion residual dropped from 0.02 mas/yr to near zero. The planet's orbit, once confidently bound, became unconstrained. Follow-up spectroscopy showed that the planet and star have different radial velocities, inconsistent with common proper motion. As of late 2024, HIP 65426b remains a directly imaged object, but its status is listed as "unconfirmed" in the NASA Exoplanet Archive.

The correction was published as a retraction note in Astronomy & Astrophysics in March 2024, with seven co-authors from three continents. The note stated that the original detection was invalid due to an astrometric artifact. It did not assign blame, but it did recommend that future studies using Gaia astrometry for planet detection should flag any solution that relies on a single high-weight epoch. The retraction had ripple effects. Several observing proposals that had been accepted based on the bound orbit were rescoped. One JWST program that planned to characterize the planet's atmosphere shifted to a different target. The ELT and Keck proposals were withdrawn. The total cost in telescope time and researcher effort is hard to quantify, but some estimates put it near 20 person-years of follow-up work that now has to be reinterpreted or abandoned.

The tension between precision and fragility

Gaia's star tracker is a marvel of engineering. It measures the spacecraft's orientation to better than 0.1 arcseconds, enabling astrometric precision that was unimaginable a decade ago. But that same precision makes the system vulnerable to minuscule disturbances. A thermal fluctuation that would have been invisible to earlier missions — like Hipparcos, which had arcsecond-level star trackers — becomes a detectable artifact in Gaia's sub-milliarcsecond regime. The very sensitivity that makes Gaia revolutionary also creates new failure modes.

The trade-off is stark: high precision comes with high fragility. The same instrument that can detect the reflex motion of a star tugged by an Earth-mass planet can also be fooled by a 30-second temperature change. This tension is not unique to Gaia. The Hubble Space Telescope's Fine Guidance Sensors, the Keck Interferometer, and the future Nancy Grace Roman Space Telescope all face similar challenges. Every improvement in measurement precision introduces a new class of systematic errors that must be understood and mitigated.

The exoplanet community has long debated the balance between discovery speed and data quality. The pressure to publish first, especially for high-profile targets like directly imaged planets, can lead to premature claims. The HIP 65426b case is a cautionary example: the signal was plausible, the statistics were marginal but acceptable, and the community was eager for a benchmark. The error was not malicious or negligent; it was a normal product of a system that prioritizes speed over thoroughness. But the episode has prompted a rethinking of how quickly results should be disseminated and how much independent verification is required before a detection is considered secure.

Statistical cost of one bad epoch

The direct cost of the error is measured in wasted resources. Roughly 20 person-years of follow-up were spent on a planet that may not exist as a bound companion. That includes the original imaging, the astrometric analysis, the orbital simulations, and the proposal writing. Three major instrument proposals — for JWST, ELT, and Keck — were rescoped or withdrawn. The opportunity cost is harder to count: those telescope hours could have been used for other science.

The false-positive rate in direct imaging is historically around 15 percent, meaning that roughly one in six directly imaged planets turns out to be a background star or a false signal. HIP 65426b joins a small but growing list of such cases. Similar astrometric revisions have affected other high-profile planets, including β Pictoris b, whose astrometric orbit was also revised after a Gaia data update. Another case, the candidate around the star HD 106906, was initially thought to be bound but later shown to be a background object based on proper motion. Each case reinforces the need for independent checks.

The broader statistical lesson is that a single bad epoch can corrupt an entire astrometric solution if it has high weight. Gaia's pipeline is designed to down-weight outliers, but the algorithm can be fooled by a measurement that is precise but inaccurate. The star tracker drift produced a measurement that was precise to 0.1 mas but offset by 3 mas — a 30-sigma outlier that the pipeline treated as a real signal because it was consistent across multiple scans within that epoch.

What the field learned from the refutation

The HIP 65426b affair has changed how the exoplanet community vets astrometric data. Several journals — including Astronomy & Astrophysics, The Astronomical Journal, and the Monthly Notices of the Royal Astronomical Society — now require that any paper using Gaia astrometry for planet detection include a statement about the number of epochs used and the weight of individual observations. The Gaia pipeline itself has been updated to flag epochs with known star tracker anomalies, and a new data release, DR4, is expected to include a quality flag for each observation.

The exoplanet community has adopted a "red flag" checklist for astrometric planet candidates. The checklist includes: (1) check that the astrometric signal is present in multiple independent epochs, (2) verify that the signal is not dominated by a single high-weight observation, (3) confirm that the proper motion residual is consistent across different Gaia data releases, and (4) obtain independent astrometry from other instruments if possible. The checklist was circulated informally after the correction and has been cited in several subsequent papers.

Open data reanalysis by third parties has become more common since the incident. The Edinburgh group that flagged the error was not part of the original team; they were doing a systematic check of all directly imaged planets. Their willingness to share the finding publicly, and the original authors' willingness to retract, set a positive example. The field is slowly learning that retractions are not failures but corrections that strengthen the literature.

The path from preprint to consensus

The correction preprint appeared on arXiv in November 2023, roughly two years after the original claim. Seven co-authors from three continents signed the retraction note, which was published in Astronomy & Astrophysics in March 2024. The timeline from first claim to consensus was about three years — fast by the standards of scientific corrections, but long enough for significant resources to be misdirected.

The NASA Exoplanet Archive now lists HIP 65426b as "unconfirmed." The object is still of interest as a possible free-floating planet or a background brown dwarf, but it is no longer considered a benchmark for atmospheric models. The original discovery paper still stands as a detection of a point source near the star, but without common proper motion, it is not a planet in the usual sense.

The process illustrates how science self-corrects, but also how slowly that correction can happen. The original claim was published in a top journal, cited by dozens of papers, and used to justify expensive observations. The retraction was handled transparently, but the damage to the field's credibility and efficiency was real. The case is a reminder that even the best data pipelines can produce artifacts, and that independent verification is not optional.

The story of HIP 65426b is not a scandal. It is a routine case of a subtle instrument error that slipped through multiple layers of quality control. The response — open data, rapid correction, and updated protocols — is exactly what the scientific method prescribes. But the episode also shows that the method works only when the community is willing to admit mistakes and change its practices. That willingness, in this case, was real. The next 0.003 arcsecond error may be caught before it causes as much damage.