Builder guide: make a workflow your own
The builder is where you change what a workflow prices and where its results go. It reads as a recipe of five stages, Price, Using, Calculate, Decide and Deliver to, beside a Preview of what the workflow produces. Test prices the window, the next 48 hours unless you change it, and delivers nothing; Publish saves your edits as the next version, the one runs use. Each control is an API call you can also make yourself, named below.
Open it
- On Automations, choose an automation, then Edit in its title row: the builder opens its newest version.
- On Automations, choose + New automation, pick a template and choose Open in the builder: it opens an unsaved copy named for the template. Publishing saves it as your workflow under the template's slug, such as
pass-yield. - A link:
/automations?view=builder&workflow=<slug>.
To keep a second copy of a template under another slug, save its graph with the API: see Recipes.
The header holds Automations (or Workflows), the way back to the list with this one chosen; the workflow's name, which you can edit in place; Draft while it has edits no version holds yet, else Published v3; Recipe and Graph; Test, which reads Publish and test while you have unpublished edits; Publish; and ⌘K, which holds every command.
The recipe
| Stage | What it holds | What you set |
|---|---|---|
| Price | What a run prices, over which window, for which part of your fleet | Passes or Links; the window, from the next 6 h to the next 72 h, this moment or the last 7 days; and the whole fleet or a part of it |
| Using | The models each pass is priced with, each adding what it says with its range | Link SNR, Weather, Demand, RF interference and SSA pass forecast, each on or off |
| Calculate | The steps that turn what the models say into what you act on, in order, such as SINR → Capacity → Pass success → Expected GB | Choose a step for what it does, what it writes, its settings and Remove step; + lists the steps that can come next |
| Decide | What the workflow recommends, and what becomes of each recommendation | Its rule as a sentence, such as "Recommend a backup pass when weather puts more than 30% of a pass at risk"; Each recommendation: Show or Send for approval; Stop recommending. With no decision yet, + Recommend changes adds one |
| Deliver to | Where each run's results go | API, always on, and a tick for each of your connections: a mission control system (MCS) whose bridge checked in, a connector you approved, a destination. Settings opens one; + Add connection opens Integrations |
What each step reads, writes and needs is in the node reference. A step is offered only once the steps it needs are in the line, and added in its stage's place, wired in, with the settings the templates give it.
The fleet button in Price opens the fleet the workflow runs on: Edit fleet, which opens Fleet; Choose a subset, to tick satellites and antennas (with everything ticked, the workflow follows your fleet as you add to it); A sample every so many seconds; and Scenario for runs, antennas down or extra fade in dB per antenna. While a scenario is set, every run is a scenario run that delivers nothing, until you clear it.
Each recommendation is the workflow's policy step: Show sends recommendations as advice, and Send for approval holds each until a person decides it on Passes. On a trial, recommendations are recorded, not sent. On a plan without Act (the part of Operator + Act that sends recommendations), Send for approval is off, each recommendation is shown as advice, and the stage says approvals need Operator + Act. See From trial to sending.
As you edit, the API checks the workflow (POST /predictions/workflows/{slug}/validate) and each problem shows in words on the stage it is about; a problem with the whole workflow shows under the header.
The preview
Before a run, the preview says what Test in the header will price. After one, it shows:
- Expected, the expected GB of every pass added up, and Could fail, the passes that deliver nothing in their low case; an audit workflow shows Delivered of possible instead. Recommendations counts any, and says whether they were sent, recorded or not sent.
- The first five results with their expected GB and range. Choose one to see its values on the graph; a link such as All 24 results (⌘J) lists every one.
- What it ran on and where it went, such as "Tested on the next 48 h · delivered nothing". After an audit run, a button beside it opens its report under Past on Passes.
The graph
Graph shows the same workflow as blocks in five columns named as the recipe's stages, Price, Using, Calculate, Decide and Deliver to, wired left to right. Each block shows what it wrote on the result you picked. Choose a block to see, beside the graph, what it does, what it writes, its value with Range and Source, its settings and Every setting, as JSON. Drag from a block's right edge to another's left edge to wire them; a condition has two exits, true and false.
The rules: one input, one API output, a model before the step that reads it, and only an output takes more than one wire (Graph rules). A workflow that branches or joins is no recipe, so it opens on the graph, says why, and Recipe is off.
Test it
- In Price, pick the window.
- Choose Test. With unsaved edits the button reads Publish and test: the API tests only a saved version, so it publishes them first.
- The header shows each stage while it runs, with Stop run. The preview fills in when it finishes.
A test run prices the window from now and delivers nothing, to no mission control system and no destination; an audit tests the last 7 days against your pass log. The API equivalent is POST /predictions with "deliver": false (reference); the builder sends an Idempotency-Key with it, so a double click starts one run. Test runs are open on every plan, except that one reading a served demand, rfi or ssa lane, or covering more than 2 satellites, needs Operator, and one with a Totals step (sum_by) needs Network.
Publish a version
Publish (⌘S) saves your edits as the workflow's next version (PUT /predictions/workflows/{slug}). A version never changes once saved, and every run names the version it ran. If someone published a newer version after you opened the workflow, the builder asks first: Open v4, or Replace it with my edits. Once published, the panel beside the recipe shows the call that reads the newest result of that version, with Copy:
curl -H "x-api-key: $CONSTELLATION_API_KEY" \
"https://api.constellation.space/predictions?workflow=pass-yield&workflow_version=3"
With nothing changed, Publish is off; ⌘S, or Show how to read v3 in ⌘K, shows the same call.
Run it
Run now (⌘↵, or in ⌘K) runs the published version over the window, and its outputs deliver. A run that delivers needs Operator or above. Stop run ends it early, and a stopped run delivers nothing. A run keeps going if you leave the page, and the builder shows it again when you come back. See Long runs.
Keys
The builder works through your sign-in session, so you need no API key of your own to test, run or publish. An account with no live key yet gets one made for it, named Platform, the first time a tab needs one; its secret is never shown.
Shortcuts
| Keys | Does |
|---|---|
| ⌘K | Every command: run, test, publish, undo, add a step, open another workflow or a template |
| ⌘↵ | Run now |
| ⌘S | Publish |
| ⌘Z, ⇧⌘Z | Undo, redo |
| ⌘J | Show all results |
| F | Fit the graph to the screen |
Related
- Node reference: every step
- The template catalog: what each template prices
- From trial to sending: what a workflow does with its recommendations
- Predictions API: workflows: the graph and run routes
Next: see what each step does in the node reference, or send results to your mission control system with Results out.