r0b: Real Or Bogus

Caution

I am actively working on this algorithm - and significant updates will be put in production throughout March 2026 at least.

The reliability score provided by Rubin is very useful for first cuts but as a transient astronomer it is not sufficient to create a clean enough stream. By using the reliability score alongside other features in the alert packet we can make better decisions. This work is based on the ATLAS Virtual Research Assistant (Stevance et al. 2025), and adds methods of Active Learning so I can train directly on the stream.

Because I am training directly on the live Rubin stream it will take a few weeks to train and verify an algorithm. The details of this will be added to this documentation and make the object of a proper paper later in the year. If you have questions in the mean-time email me or start a topic on the Lasair Community Forum.

The r0b LVRA is a binary classifier downstream of a filter that makes preliminary cuts: * reliability >0.5 * SHERLOCK label either SN, NT, ORPHAN or UNCLEAR. * nDiaSources` >= 2 *

Warning

The r0b scores are absolutely not suitable for Solar System Objects of AGNs. If they work well, it is on accident. Use at your own peril.

How does it work? What data was it trained on? Performance?

Come back in March/April 2026