AstroOracle

Active learning oracle for astronomical anomaly triage.

• Fetch SkyView cutouts (or synthetic offline cutouts)
• Rank candidates via anomaly score + acquisition + diversity
• Label in CLI or in a small web UI
• Retrain hook when enough new labels are collected

Badges

• CI: GitHub Actions (see .github/workflows/ci.yml)
• Release: GitHub Actions tags v* (see .github/workflows/release.yml)

Install

pip install -e ".[dev,api,plotly,graph]"

Optional extras:

pip install -e ".[astro,explain,docs]"

Quickstart

1) Web landing page (FastAPI)

astrooracle serve --candidates candidates.parquet --annotations annotations.csv
# then open http://127.0.0.1:8000/

Landing page features:

• GET / landing page with Plotly 3D preview and a CLI-like /run form
• GET /dashboard annotations dashboard + export
• GET /viz3d Plotly JSON
• GET /graph_anomaly graph context (kNN on sphere)
• GET /explain lightweight explanations + shareable prompts

2) CLI triage (interactive)

astrooracle run --candidates candidates.parquet --annotations annotations.csv

3) Generate a static batch annotator (HTML)

astrooracle batch-html --out-dir batch_out --n-query 12
# open batch_out/index.html

New features included in this patch

• Landing page / + dashboard /dashboard (Bootstrap + Plotly)
• Toggle "Mode Chaos" (served via /chaos, requires a timeseries column)
• Graph anomaly context (astrooracle graph-anomaly or /graph_anomaly)
• Explainability JSONL (astrooracle explain-top or /explain)
• Gaia ingestion (astrooracle gaia-cone and astrooracle gaia-adql, requires .[astro])
• Hybrid anomaly fusion (--mode hybrid in CLI and landing page)

Citation

See CITATION.cff.

License

MIT.