Argus

Telescopes watch the same patches of sky, night after night. Most objects behave predictably — the same brightness, the same rhythms, over and over. Some don't: every so often, an object's brightness changes in ways the usual models can't explain. Argus watches public sky-survey data and flags those objects for a human to review.

Open the sky → GitHub repo

What Argus is

Argus pulls light-curve data on astronomical objects from the public ZTF survey (via the ALeRCE broker) and runs a few simple comparison checks on each one. When the simple checks don't fit cleanly, the object is queued for a human reviewer.

The thing Argus produces is not a score, and not a classification. It's a small case file for each flagged object: observed brightness over time, where simple models match and where they don't, recommended next checks, and the uncertainty and caveats around all of it.

The web app is just a way to read those case files. The sky view is the entry point; a story view shows one object at a time in plain language; an Expert view exposes the full evidence panels for anyone who wants the precise terms and numbers.

A look

Argus sky view — full-screen Aladin sky with the overlay sentence
The sky view is the entry point. The first-visit sentence fades after a moment.
Argus story view — one flagged object in plain language
The story view explains one flagged object in plain English, with brightness over time as the hero.

Guided tour

  1. Open the workstation at greg-oyan.github.io/argus/workstation/. You arrive inside a sky view with flagged objects marked.
  2. Click a marker. The view flies in and opens that object's story page.
  3. Read the three plain-English questions: what is this, why was it flagged, what would an astronomer check next. All three are derived from the case-file JSON; nothing is invented.
  4. If you want the technical layer, open the Expert view. It contains the full evidence panels (assessment, residuals, comparators, features, narrative, sky context, next checks), the priority-ordered queue table, and a short glossary.
  5. If you'd rather read static reports, every case has portable HTML, Markdown, and JSON exports — links live under the public review queue.

Honest boundaries

Argus is a case-file-first astronomy review aid, not another classifier. These examples do not decide any object's physical identity or assert a final finding. They organize evidence for further review.

Broker labels and catalog matches are metadata, not conclusions. Review priority is a queue sorting heuristic, not a model result. The anomaly_assessment field is an evidence triage summary inside a case file, not an object-identity claim.