Scheduling & repetition

When you set up an observation you describe more than what to capture. You also describe how it repeats, how it is grouped in time, when it runs, and which telescope takes it. These controls live on the Configuration step (plus a per-request count on the Data Requests step). This page explains what each one means and how Skynet acts on it.


How an observation breaks down

Every observation is built from a few nested pieces. Understanding them makes every setting on this page intuitive.

  • Observation — the whole thing you submit.
  • Epoch (a visit) — one complete pass over your target. "Come back each night for a week" is 7 epochs.
  • Request — within a visit, each request is one kind of measurement taken to a chosen depth — for example one filter of an optical image, or one frequency setup of a radio spectrum. An observation can have several requests (call them A, B, C), each with its own settings.
  • Sample — a repeated measurement of one request within a single visit. "10 samples of request A" means ten A measurements in the same visit. Samples give you temporal resolution — they let you watch something change during a visit.
  • Frame — the actual capture(s) the telescope takes. To reach the depth you asked for, the telescope may split one sample into several frames, deciding how many from the conditions on the night. You never count frames — you ask for a depth, and the telescope does the rest.

Within a single visit, your requests and their samples form a grid. With three requests (A, B, C) and 10 samples each:

               sample 1   sample 2   …   sample 10
    A   →         A1         A2       …      A10
    B   →         B1         B2       …      B10
    C   →         C1         C2       …      C10

One more term names a useful slice of that grid:

  • Segment — the contiguous mid-tier chunk of a visit. Depending on the order you take your measurements (see Sample ordering below), a segment is either a single request's whole run — one row, all ten A samples — or one sample of every request taken as a set — one column, one A, one B, one C.

So: each cell is a sample, a row or column is a segment, the whole grid is one epoch, and repeating the grid over time gives you more epochs.


Depth — how faint you go

Depth is not a repeat count. You set it per request (on the Data Requests step) as a target signal-to-noise against a brightness estimate. At the telescope, that target is turned into the right total exposure time for the night's conditions and split into frames automatically. Two requests can ask for the same depth and end up with different numbers of frames on different nights — that is expected and handled for you.

The settings below control repetition, grouping, and timing; depth controls how deep each sample goes.


Repetition: epochs and samples

Setting Where What it controls
Epochs (visits) Configuration How many times to revisit the whole target. Set 0 to repeat indefinitely until you cancel.
Samples each request (Data Requests) How many measurements of that request to take per visit.

These two are independent. Epochs are revisits of the whole target over time; samples are repeats of one request within a single visit. A week-long nightly campaign with three quick measurements of each request per night is 7 epochs with 3 samples on each request.


Keep together (cohesion)

Cohesion is the "in one go" control: it names the largest unit that must run without interruption. Everything coarser than that unit may have gaps; nothing inside it will be broken up.

Choice Meaning
Sample Each individual measurement runs in one piece; the schedule may pause between samples.
Segment A run within a visit stays together. Its shape comes from your sample ordering: either all samples of one request (a request's whole run), or one sample of every request taken as a set.
Epoch (the whole visit) The entire visit — every request and sample — runs uninterrupted.
Observation (every visit) The whole observation, across all visits, runs as one uninterrupted block.

Choose a finer unit for flexibility (the scheduler can fit your pieces in wherever there's time) and a coarser unit when the science needs an unbroken run — for example, an uninterrupted run of one request for a clean combined result, or a whole visit held together for a fast-changing source.

Some combinations don't make sense together — for instance, requiring the whole observation to run uninterrupted while also asking for one visit per night. The editor keeps your choices consistent and won't let you build a contradiction.


Sample ordering

When a visit has more than one request and more than one sample, ordering decides the order the measurements are taken within the visit:

  • Finish each request before the next — take all of one request's samples, then move on. A clean, contiguous run per request.
  • Take one of each request, then repeat — take one sample of every request, then come back for the next round. The measurements from each round land close together in time.

This is also what gives a Segment its shape: with finish each request first, a segment is one request's whole run; with one of each, then repeat, a segment is one round across all requests.


Cadence — spacing in time

Cadence controls the time between repeats:

Choice What it does
As soon as possible No imposed spacing — repeats run whenever conditions allow.
Once per calendar night / day At most one visit per night (or per day), spreading a multi-epoch campaign across successive nights.
Minimum gap between units Records a smallest spacing you'd like between repeats.
Fixed period between units Records a regular interval you'd like between repeats.

The scheduler imposes spacing for as soon as possible (none) and once per night/day (one per calendar period). A minimum gap or fixed period is saved with your observation as the spacing you intend; the scheduler still places each unit as soon as it is observable rather than holding it back to enforce the interval.


Copies

Copies decide how many independent datasets you get:

Choice What it does
Single dataset One copy of the data, on one suitable instrument.
All eligible instruments Mirror the work to every instrument your grants can reach — one copy per instrument.
A fixed number of instruments Mirror to a set number of instruments.

When you ask for more than one copy, coordination appears:

  • Independent — each copy is scheduled on its own, whenever its instrument is free.
  • Synchronized start — records that the copies should begin together.

Each copy is a full, separate dataset, so asking for several copies uses your observing time accordingly.


Instrument assignment

This controls whether a single instrument is tied to the whole observation or may change as it runs:

  • One instrument for the whole observation — every visit and request is taken on the same instrument.
  • May vary per request / visit — the scheduler may place different requests (or visits) on different eligible instruments, picking the best fit for each.

When you allow the instrument to change, a handoff timeout is recorded — how long an interrupted piece should wait before it can be reassigned to another instrument.


On interruption

If a unit you asked to keep together is interrupted (weather, say), this records what should happen when it resumes:

  • Resume from where it stopped — keep the partial work.
  • Restart the interrupted unit — start that unit over.

Defaults

If you change nothing, an observation is a single deep visit by one telescope:

Setting Default
Epochs 1
Samples (per request) 1
Keep together the whole visit (epoch)
Sample ordering finish each request before the next
Cadence as soon as possible
Copies single dataset
Instrument one for the whole observation
On interruption resume

Worked examples

A single deep image. Leave the defaults. One visit; each request taken once to your target depth; the telescope splits each sample into however many frames the night's conditions require.

Nightly monitoring for a week. Set Epochs = 7 and Cadence = once per night. You get one visit per night for seven nights, each a complete pass over the target.

A fast time series of one request. On a request, set Samples = 30, set Keep together = Segment, and set Sample ordering = Finish each request before the next. The 30 measurements of that request run back-to-back as one uninterrupted, evenly sampled series.

Near-simultaneous across requests. With several requests, set Keep together = Segment and Sample ordering = Take one of each request, then repeat. Each round — one sample of every request — is taken as a group, so the requests are measured close together in time, repeated each round.


← Step 4 — Configuration · The observation spec →