Eclipsing binaries at the conjunctions predicted by near-edge-on Gaia DR3 astrometric orbits: data, predictions and code

Data, code and predictions for the Research Note of the AAS "Eclipsing binaries at the conjunctions predicted by near-edge-on Gaia DR3 astrometric orbits" (Chopard-Vilhem 2026). Unlike spectroscopic orbits, Gaia DR3 astrometric orbits give the inclination, and an orbit seen edge-on predicts when its eclipses occur. The 1726 astrometric orbits brighter than G = 14 whose best-fitting inclination lies within 1° of 90° were selected, their uncertainties propagated to the times of the two conjunctions, and the TESS light curves (SPOC, TESS-SPOC, QLP) searched at those times. A classification rule, frozen before the joint statistic was computed, compares each system's consistent eclipses with simulations at the false-alarm rates of a face-on control and of a pessimistic case: three new eclipsing binaries are established (Wolf 1392, HD 145453, TYC 3908-665-1), three are candidates (HD 284424, HD 170829, TOI-1656), and one is not retained (CD−81 705). Contents the selection of 1726 orbits and their Gaia DR3 orbital elements; every search window and flag, all single events, and the false-alarm trials; the face-on control: the same search on 3183 orbits over 30° from edge-on, which cannot eclipse, with the commands and seeds that produced it and the ordered list of candidate orbits from which both draws can be checked; the classification rule (its exact text, SHA-256 and the commit that froze it), the bootstrap uncertainties of every flagged dip, the Gaia period errors, and the result of the rule for every system and control group; the artifact vetting and its calibration on 256 eclipses of eclipsing binaries found from Gaia spectroscopic orbits; the injection–recovery tests and the expected yield of the search; pixel-level difference images and literature checks of the candidates; predicted times of the next eclipses with their uncertainties (October 2026 to June 2028), and a calendar of those observable from Cluj-Napoca; the figure of the note and all the code. Version 1.4.0 is the version cited by the note; it supersedes versions 1.1.0 to 1.3.0. It adds a classification rule written, hashed and committed on 3 November 2026 before the joint statistic was computed (classification_rule.md). For each system, K is the size of the largest subset of its flagged dips consistent with one period and one epoch per conjunction, in time and depth; its probability comes from a joint simulation of the system's windows at two reference rates per pipeline: (a) the face-on control, (b) a pessimistic case built from the search itself. Besides the hypotheses drawn from pairs of dips, the code also tries the Gaia period itself, as declared at the head of the rule's code block; this can only increase K, in the data and in the simulations alike. Established: P ≤ 0.01 under (b); candidate: otherwise P ≤ 0.05 under (a), with at least two consistent dips. Of the two catalogued eclipsing binaries used as positive controls, HD 75767 is established and KOI-3578, observed in QLP only, is not retained; 1 of the 12 control stars with at least three flagged dips reaches P ≤ 0.05 under (a), and none is established. The passages of the note that rested on the propagated timing uncertainties are removed; the figure (version 6) shows the classification and only the dips retained by the rule; code/check_numbers.py checks the 53 numbers of the note against n2_numbers.txt. See the README for details. Reproducibility. Every number quoted in the note is recomputed from these files by code/n2_statistics.py into n2_numbers.txt ("N2" is the working name of the note), with fixed random seeds; for this version this was reproduced identically from a fresh copy of the archive on the author's Mac (Python 3.13, and a fresh Python 3.12 environment) and on a Linux x86_64 machine (Python 3.12.3, numpy 2.4.4, pandas 3.0.2, scipy 1.17.1); only least-squares values differ, in their eleventh significant digit. See README.md for the contents of each file and notes for re-use. Testing the predictions. predictions/eclipse_predictions.csv gives the mid-eclipse times (UTC), their 1σ uncertainties, depths and durations. Only eclipses after the publication date of version 1.0.0 (13 October 2026) are genuine forecasts. Observations that confirm or refute them are welcome: [e-mail]. License. Data, predictions and figure: CC BY 4.0. Code: MIT (see LICENSE in the archive).