Skip to content

Alert subscriptions

An alert subscription tells Skynet: “when an external astronomical alert matches these conditions, create this observation automatically.” When a matching alert arrives — a gamma-ray burst, a gravitational-wave detection, a survey transient — Skynet builds a pre-configured observation with the event’s position filled in and hands it to the scheduler, with no human in the loop.

Skynet maintains the connections to the alert brokers centrally, so you never configure a feed or a credential. You just describe what you care about and what to do about it.

Typical uses:

  • Time-domain follow-up — image the position of every new transient that passes your filter (a nearby supernova candidate, a kilonova, a bright GRB).
  • Coordinated campaigns — point an organization’s telescopes at the same events as part of a follow-up consortium.
  • Education — surface high-impact events to a classroom account.

Every subscription has three parts:

  1. A source and a filter — which alert stream to listen to, and which alerts within it are worth acting on.
  2. A response observation — the observation to create when an alert passes the filter. It’s a saved observation spec with the target left open; Skynet fills in the alert’s position when it fires.
  3. Funding and policies — the observing grants that pay for fired observations, plus rules for how often to fire and what to do when an event is later updated or retracted.

Alerts arrive from two brokers. Each subscription listens to one stream; to cover several missions, create one subscription per mission (each gets its own rate limits).

NASA GCN (the Gamma-ray Coordinates Network) delivers rapid positions for high-energy transients:

  • Swift gamma-ray bursts — burst positions from Swift’s BAT, XRT and UVOT.
  • Einstein Probe — fast X-ray transients from its Wide-field X-ray Telescope (WXT).
  • Gravitational-wave alerts (LVK) — LIGO/Virgo/KAGRA compact-binary and burst candidates.

Fink delivers classified optical transients from time-domain surveys (such as ZTF), enriched with machine-learning classifiers:

  • Kilonova candidates and early supernova candidates (both from ZTF).

Open Settings → Alerts on your account (or an organization you manage) and choose New subscription. The editor is organized into a few sections.

  • Name — a label for the subscription.
  • Alert source — one of the streams above.
  • Fire on notice types — whether to act on the initial notice for an event, later updates (refined positions the event hasn’t fired on yet), or both. Most subscriptions fire on the initial notice.
  • Enabled — a new subscription starts enabled; Disable / Enable in the subscription list toggles it without deleting it.

Filters are how you narrow a firehose of alerts down to the handful you want to chase. Leave a filter blank and it doesn’t restrict anything — you opt in to each cut you care about. The exception is event age, which defaults to six hours. The filters available depend on the sources you picked.

General (any source):

  • Localization size — reject events with a poorly-constrained position (a large error region), which are hard to image.
  • Declination band and galactic-latitude cut — keep events your telescopes can actually reach, or restrict to extragalactic sky.
  • Moon and Sun separation — skip events too close to the Moon or Sun to observe well.
  • Event age — ignore notices older than this when they arrive (6 hours by default; useful protection against backfilled or replayed streams).
  • Active months — only fire during part of the year (e.g. your observing season). Set through the API only; the editor does not show it.

Gamma-ray / X-ray bursts (Swift, Einstein Probe):

  • Instruments — restrict to specific mission instruments (API only).
  • Minimum significance (SNR) — only bright-enough detections.
  • Exclude bogus alerts — drop known false-trigger flags (on by default).

Gravitational waves (LVK):

  • Significant only — ignore low-confidence candidates.
  • False-alarm rate — require a candidate rare enough to be worth following (the canonical electromagnetic-follow-up cut).
  • Classification — minimum probabilities for BNS, NSBH and has-NS. Restricting to compact-binary vs. burst candidates and a has-remnant threshold are available through the API only.
  • Distance and 90% sky-map area caps — keep events near enough and well-enough localized to tile.

Survey transients (Fink):

  • Kilonova / supernova score — minimum classifier confidence.
  • Brightness — a magnitude limit.
  • Age and number of detections — how new and how well-sampled the object is.

Attach one or more response observations. Each is a saved observation spec — the easiest way to make one is to build the observation you’d want in the editor, then import it here from that observation or export its spec and paste it. For each response you set:

  • A name template for the fired observation, with the placeholders {event_name}, {external_id}, {source} and {event_time} (UTC) — e.g. GRB {event_name} follow-up.
  • An optional start delay after the event time.

When an alert fires, the spec’s target is replaced with the alert’s position and the observation is created. (If an alert has no position, that fire is skipped.) Attaching several responses lets one alert kick off several observations — say a rapid wide-field image plus a deeper follow-up.

  • Observing grants (required) — fired observations are charged against these, exactly like observations you create by hand. If every attached grant is later revoked (or its account no longer belongs to the subscription’s owner), the subscription disables itself rather than firing un-funded. No notification reaches you when that happens, so check the subscription if alerts stop producing observations.
  • Instruments — there is no separate instrument setting; fired observations use the instrument restrictions saved in the response spec.

Events evolve — positions get refined, and some are retracted as false alarms. You choose what happens after the first fire:

  • On position update — retarget the existing observation to the refined position (the default), retrigger a new observation, or ignore the update.
  • On retraction — cancel the fired observation (the default), notify only, or ignore the retraction.

Guards against a noisy stream flooding your night:

  • Max fires per night — cap the number of fires in a rolling 24 hours.
  • Cooldown — a minimum gap between consecutive fires.
  • Auto-cancel after — automatically cancel a fired observation a set number of hours after the event, so stale follow-ups clean themselves up.
  • Expires on — an optional date after which the subscription stops matching. Set through the API only; the editor does not show it.

Two tools help you trust a subscription before it runs live:

  • Backtest. Run your draft filters against the alerts Skynet has actually received over the last 90 days (up to 200 of them). You’ll see how many would have matched and which filters rejected the rest — so you can tune the thresholds before turning the subscription on.
  • Test fire. Turn on Fire on test alerts in the subscription’s Testing section, then press Test fire in the subscription list to send a synthetic alert and watch the full path execute — funding check, target substitution, and observation creation — without waiting for a real event. The website fires at RA 180°, Dec 0°; the API lets you choose the position.

Settings → Alerts has three views:

  • Subscriptions — your subscriptions, each with quick enable/disable, edit, and delete.
  • Activity — a log of every match decision: what fired, what was skipped (and why — rate-limited, no position, filtered out), and a link to the observation that was created.
  • Alert stream — the recent alerts Skynet has received across all brokers, so you can see what’s flowing in.

A fired observation is a normal observation owned by the same account or organization as the subscription. It shows up in your observation list, runs against your grants, and is yours to edit, watch, or cancel like any other — see Observation detail.