Yahalomi et al. release simulated Gaia DR4 epoch astrometry for 15,120,984 injected systems
The dataset injects zero, one and two companions into 5,040,328 main-sequence stars within 500 pc, with known parameters for testing orbit-modeling pipelines.
Yahalomi et al. (2026) published, via doi.org, a record of simulated Gaia DR4-like epoch astrometry. It is meant for testing orbit-modeling pipelines on a common set of injected systems with known parameters 
doi.orgdoi.orgdoi.orgOpen source ↗. Into each of the 5,040,328 main-sequence Gaia stars within 500 pc in the G23H catalog, the authors injected a zero-, one- and two-companion system, for 15,120,984 systems in total doi.orgdoi.orgdoi.orgOpen source ↗.
Injection and detection setup
Companion masses and semimajor axes are drawn log-uniformly over 1 M_Mars to 80 M_Jup and 0.03 to 100 au doi.orgdoi.orgdoi.orgOpen source ↗. The set uses no occurrence model and no detectability cut, so the companion-free population serves as a matched null doi.orgdoi.orgdoi.orgOpen source ↗. The one- and two-companion populations each carry a high-SNR sub-catalog, in which every companion has SNR_tot of at least 5. These contain 439,503 and 26,379 systems, respectively doi.orgdoi.orgdoi.orgOpen source ↗.
Every system was characterized with the kepmodel astrometric periodogram on a grid of 15,777 trial periods doi.orgdoi.orgdoi.orgOpen source ↗. The detection thresholds were calibrated on the companion-free population to a 1% false-positive rate and applied to all three populations doi.orgdoi.orgdoi.orgOpen source ↗. The characterization tables give the five-parameter chi2, best period, class (undetected, unimodal, multimodal or broad), peak powers, delta-BIC, acceleration test, calibrated detected and period_reliable flags, and FAP doi.orgdoi.orgdoi.orgOpen source ↗.
Files and tools
The stellar sample, stars.csv, holds G23H astrometry, Gaia DR3 photometry and astrometric metadata, Hipparcos-Gaia proper motions, per-star noise-model terms, and interpolated main-sequence mass and radius doi.orgdoi.orgdoi.orgOpen source ↗. The injected truth for every system is stored in injected_solutions parquet files. For each companion they list sma, ecc, mass, inc, Omega, omega, M_anom, period and several signal-to-noise values doi.orgdoi.orgdoi.orgOpen source ↗.
Each population has 640 parquet shards: 320 epoch shards and 320 matching truth shards. An epoch row contains the observation time, scan position angle, parallax factor, along-scan centroid position and its error, and the field of view doi.orgdoi.orgdoi.orgOpen source ↗.
The generator, the periodogram characterization, the notebooks for the paper figures and an orbit-fitting tutorial based on NumPyro and jaxoplanet are available at github.com/dyahalomi/epochalypse doi.orgdoi.orgdoi.orgOpen source ↗. The script download_catalog.sh fetches the record into the expected layout and unpacks the epoch shards on request. MD5SUMS.txt lists an MD5 for every file doi.orgdoi.orgdoi.orgOpen source ↗.
