DRAFT
2026-04-21 16:03:00
Type: Announcement-Tool/Utility
Bibcode: 2026TNSAN.107....1S
The Lasair Virtual Research Assistant (LVRA) annotator for Rubin alerts - r0b.a1
Authors: H. F. Stevance, S. J. Smartt (Oxford/QUB)
Abstract:
We present the first prototype of the Lasair Virtual Research Assistant (r0b.a1), a publicly available annotator on the Lasair Rubin Data broker which has been annotating the alert stream since the first public alerts on 24 February 2026. It scores alerts between 0 and 1 to provide better ranked filtering of the Rubin data stream where the main aim is to find extra-galactic transients. A paper is in preparation but the annotator is already being publicly used by 60 community filters so here we briefly describe the annotator, usage recommendations, caveats and methodology.

Lasair Annotator 

The r0b annotator scores the Vera C. Rubin Observatoryalerts on the Lasair broker  (Williams et al. 2024) between 0 (bogus) and 1 (real extragalactic). These alerts have been pre-emptively processed by Sherlock (Young et al. 2023) and given a flag of SN (supernova), NT (nuclear transient), ORPHAN or UNCLEAR. This excludes Variable Stars, Bright Stars and AGNs and any Lasair filter using r0b will therefore exclude these categories as well.  The annotator also provides flags that may be useful for filtering transients in Lasair:

  • n_gt22: Number of light-curve points brighter than 22nd magnitude (Also n_gt21, n_gt20, n_gt19, n_gt18)
  • brighter22: Whether the current light-curve point (diaSourceId) is brighter than 22nd magnitude (same for 21st, 20th, 19th, 18th mag)
  • first22: Whether it is the first time this object has been brighter than 22nd magnitude  (same for 21st, 20th, 19th, 18th mag)
  • blame_diaSourceId: Which diaSourceId is responsible for the r0b score (currently the latest)

Following community feedback, we are working towards additional annotator features recording the history of the r0b score for a given objects. Additional feedback on the r0b annotator classifications (scores) or the data provided are welcome and you can open an issue on the LVRA GitHub

Living documentation will be maintained on the lasair VRAs read-the-docs: https://lasairvras.github.io/lvra-doc/intro.html

 

Scoring and recommended thresholds

Work is underway to characterise the performance and retrain the annotator to improve rankings, but preliminary analysis shows that selecting alerts with score >0.9 selects the top 25% of alerts delivering a purity of 75% extra-galactic alerts. Down to scores of 0.75, the r0b annotator provides higher purity than ranking by the latest R (reliability) score. Below this r0b and the reliability score provide similar extra-galctic purity but with a different mixture of contaminants (AGNs, movers, bogus). 

 

Method

The Lasair Virtual Research Assistant (LVRA) methodology is based on the ATLAS VRA (Stevance et al. 2025). The key differences are that the LVRA was trained using an Active Learning approach directly on the Lasair data stream with 411 samples received during commissioning between December 2025 and January 2026, selected to be the most informative using a bespoke algorithm that will be described in the upcoming paper. 

 

Ask Questions / Feedback

Open an issue on the LVRA GitHubor email hfstevance@gmail.com 

HFStevance and the VRA project are supported by Schmidt Sciences. Lasair is supported by the UKRI Science and Technology Facilities Council as part of the LSST:UK Science Centre and is a collaboration between the University of Edinburgh, Queen’s University Belfast and the University of Oxford, funded by grants ST/X001334/1 and ST/X001253/1. The NSF–DOE Vera C. Rubin Observatory, funded by the U.S. National Science Foundation and the U.S. Department of Energy's Office of Science, will perform the Legacy Survey of Space and Time using the LSST Camera and the Simonyi Survey Telescope. The Rubin Observatory is a joint Program of NSF NOIRLab and DOE’s SLAC National Accelerator Laboratory.

Comments