Workflows
A workflow answers one operational question for every pass in your fleet: how many GB each pass will deliver, which antennas run out of capacity, which passes rain will cost and what to book instead. It is a graph of steps you save under a name, run over a window, and send wherever your operators already look.
Expected GB comes as a range: p10 is the low case (nine passes in ten deliver at least that), p50 the median and p90 the high case. Every result carries its ledger: the value each step gave it, with the unit and where it came from. A forecast you cannot trace is not one you should act on.
What a workflow is
A workflow is a named graph, saved as numbered versions under a slug you choose. A run takes the passes (or links) in a window and moves each one through the graph:
| Step | What it does | Examples |
|---|---|---|
| Input | One row per pass, or per link, over the window | passes, links |
| Lane | What one model or physics says about each row | snr, weather, demand, rfi, ssa |
| Transform | Combines lanes into the numbers you act on | sinr, capacity, expected_gb, sum_by, swap, ledger |
| Condition | Splits rows on a threshold | threshold |
| Output | Where the results go | /predictions (always), a mission control connector, a destination |
The graph shape, each step's parameters and the run API are in the API reference.
The builder
The builder opens from the Automations tab: choose an automation, then Edit in its title row, or start one with + New automation, which offers the templates. A link to one automation's builder is /automations?view=builder&workflow=<slug>.
The builder reads a workflow as a recipe of five stages, each a sentence you edit in place:
- Price: what a run prices (passes or links), over which window, for which part of your fleet. A scenario for runs, antennas down or extra fade, is set here too; a scenario run delivers nothing.
- Using: the models each pass is priced with, one toggle each: Link SNR, Weather, Demand, RF interference and SSA pass forecast.
- Calculate: the steps in order, such as SINR, Capacity, Pass success and Expected GB. A step opens its settings; + offers the steps that fit next, the likely one first, added with working settings and wired in.
- Decide: shown when the workflow recommends a change, such as a backup pass under rain: when to recommend, and whether each recommendation is shown as advice or sent for approval. On a trial, nothing is sent: each recommendation is recorded, so you can see what would have gone out. On a plan without Act (the part of Operator + Act that sends recommendations), each recommendation is shown as advice, and approvals need Operator + Act.
- Deliver to: the API always, then your own connections as toggles: mission control systems whose bridge checked in, connectors you approved, and your destinations. + Add connection opens Integrations.
The check runs as you edit and puts each problem on the stage it is about. Graph shows the same workflow on a canvas, one column per stage, to inspect it or for a shape the recipe cannot draw; a workflow that branches opens there and says why. Select a step on the graph to see what it does, what it writes, its value on a sample pass and its settings; select a lane to see which of your satellites and ground stations it cannot price yet, with Fix in Fleet.
- Test prices the window with the published version and delivers nothing; with unpublished edits it publishes them first. The preview beside the recipe shows what the workflow produces: expected GB, passes that could fail in their low case, recommendations, the first passes, and what it ran on.
- Publish saves your edits as the workflow's next version. The API checks the whole graph first.
- Run now (⌘↵, or ⌘K) runs the published version over the window and delivers to every output.
- ⌘K holds every command: run, undo and redo, tidy the graph, add any step, open another workflow or start from a template.
The builder signs in with a session made from your sign-in, which may edit, publish and run; it needs no API key of yours.
Workflows run on your network config. On an account without one, Fleet, Passes (the audit under Past too), Scenarios on Forecasts and the builder offer Load the demo fleet: 5 ground stations, 7 antennas and 6 Earth observation satellites (5 sun-synchronous and 1 geostationary), with simulated telemetry, ready to run on. Or choose Start from the catalog to pick your own satellites and stations from the public asset catalog. Replace either any time. The API equivalent is POST /topology/demo-fleet.
Every lane can price every demo satellite: all five fly sun-synchronous orbits at 700 to 830 km, inside the 400 to 2,000 km ssa has measured orbit errors for, with passes for demand to share out. The Fleet tab shows which lanes can price each satellite and station.
Templates
Templates are read-only workflows. Opening one in the builder, or saving its graph under your own slug, gives you a copy to change.
| Template | Answers |
|---|---|
pass-yield | Expected GB for every pass, with its ledger, best passes first |
antenna-capacity | Sellable minutes and expected GB per antenna per UTC day |
modcod-plan | A time-tagged MODCOD schedule per pass, from its SINR at each step |
rain-fade-swap | Passes where weather, at its p50, costs more than 30% of the clear-sky GB, with a backup pass to book before the cancel deadline |
audit-ledger | Each pass over a past window, against the contacts you logged |
rfi-watch | Links with RF interference past a threshold, sampled every 60 s over the next hour, with their lock-loss risk and a reroute to suggest |
A template that reads a lane no model serves on your deployment is listed with available: false, and the template picker dims it and marks the lane not served. It cannot run there: the API refuses it with 422 lanes_not_served. The template catalog says what each one prices.
Runs and test runs
| Run | Test run | |
|---|---|---|
| In the builder | Run now (⌘↵) over the window | Test over the window |
| In the API | POST /predictions | POST /predictions with deliver: false |
| Computes and stores items | Yes | Yes |
| Publishes to mission control connectors and sends to destinations | Yes | No |
| Plan | Operator or above | Every plan |
A test run answers exactly what a run would, so you can check a workflow before anything reaches an operator. Explore's runs are all test runs. They ask every served model. Read either kind with GET /predictions?run_id= until its status is done.
Send an Idempotency-Key header with POST /predictions so a retry or a double click starts one run: the same key and request within 30 days answers 200 with the first run's run_id, and the same key with a different request 409 idempotency_key_reused. The builder sends one for you.
GET /predictions?workflow=<slug> reads the newest finished run of the workflow's current version. A read never starts a run and takes no window; with no finished run yet it answers 409 no_finished_run, so start one with POST /predictions.
A run that reads a model lane (demand, rfi or ssa) is sized when you save it and when it starts. One with more satellites x antennas x steps than a passes listing computes is refused with 422 too_many_link_steps, and one that would ask a model more than 1,000 times (once per pass for demand and ssa, once per visible step for rfi) with 422 too_many_model_lane_requests: name fewer satellites or antennas, or set a shorter window or a longer step. The Platform's own passes reads split such a window for you.
A run keeps going on the server until it finishes or someone stops it. Leaving the builder, reloading, closing the laptop or signing in on another device does not stop it: while it runs, the Automations tab shows a dot, other pages show it in the header, and the builder shows its progress when you open the workflow again. To end it early, press Stop in the builder's header and confirm. A stopped run delivers nothing; one already publishing or sending its results cannot be stopped. In the API, each workflow's active_run names its run in progress, and POST /predictions/runs/{run_id}/cancel stops it. When a newer run of the same workflow has finished or is delivering first, the older one sends nothing and its result says superseded_by; the builder says so instead of showing a failed delivery.
Upcoming on Passes reads the newest run of a workflow that prices every pass, such as pass-yield, and lays your other workflows over it. Past on Passes, with Audit chosen, uploads a pass log and runs audit-ledger over it; with no audit workflow yet, it creates one from the template in one click. The audit needs a fleet first, since a pass log is matched against your satellites and antennas, and on the demo fleet it offers a sample pass log in one click. The tabs, the board and the audit sign in with a session made from your sign-in, renewed for you, so they need no API key of yours and a reload or another device needs none pasted. A disabled account gets no session, and each tab says so. See Free audit.
Scenarios
A scenario is a what-if overlay on your topology config: antennas or stations taken down, antennas or satellites added, the weather lane held at its p50 or p90, or extra fade in dB on an antenna. The extra fade adds to the weather lane's fade; it does not replace it. Create one with POST /topology/scenarios, or under Scenarios on Forecasts, then run a workflow against it.
A scenario run never delivers. Whatever you send for deliver, it publishes to no connector and sends to no destination, so a what-if can never reach an operator as a forecast.
Because it never delivers, a scenario run is open to every plan, like a test run. On Explore, both run over at most 2 satellites of your own (the demo fleet's never count); name them with the input's satellite_ids when your config holds more. In the Platform, Passes asks which 2 and saves them for you; see Price the window. The demand, rfi and ssa models price your stored network config, so a run with one of those lanes takes only a scenario's weather changes: a scenario that takes down antennas or stations, or adds antennas or satellites, is refused with 422 scenario_changes_network_for_model_lanes.
Plans and limits
Listing, saving, validating and test-running a workflow is open to every plan, with the test run limits above and one more: a graph with a sum_by step, as in antenna-capacity, needs Network to save or to run, test runs included, and validating it reports plan_required. The rest is gated by plan:
| Plan | Adds |
|---|---|
| Explore | Test runs on every model, over at most 2 satellites, and a free 48-hour pass audit |
| Operator | Runs that deliver, calibration, your whole fleet, and outputs to Yamcs, Open MCT, your own connectors, Amazon S3, Grafana Cloud and webhooks |
| Operator + Act | A run's actions, such as a swap or reroute to advise, sent with its outputs (below it, a run keeps them as what Act would have done), and approving held actions |
| Network | The antenna-capacity template |
| Enterprise | Outputs to Anduril Lattice and OpenC3 COSMOS |
Each plan includes everything above it. A run that needs a higher plan is refused with 403 plan_required, whose plan and feature say what it needs (standard is Operator, pro Operator + Act, scale Network). Validating a graph reports the same thing as an issue on the step that needs it. A template's plan applies to saving its graph and to test runs too.
A run whose stored items have expired answers 410 run_data_expired; run the workflow again. Constellation keeps each run's workflow results for connectors for 7 days.
Send results to mission control
Each output on a workflow is a place its results go:
- A mission control output (
ref: "feed"in the API) for Yamcs, OpenC3 COSMOS, Anduril Lattice or your own system. A connector in your network pulls it; nothing connects into your network. See Connectors. A bridge starting for the first time readsGET /predictions/feed?from=latest: no entries, and a cursor just before the newest run, so it applies that run and none older. - A destination output for Amazon S3, Grafana Cloud or a webhook. Constellation pushes each run there.
Send for approval
In the builder's Decide stage, set Each recommendation to Send for approval (a policy step in approve mode), and each recommendation waits for a person instead of going out as advice. Deciding one says what a decision does and what it needs, and From trial to sending how a recommendation reaches your mission control system. Decide in the API with /predictions/actions, or ask an agent over MCP.
Decide in the Platform
Held actions are recommendations in the Platform: those waiting on you are under Upcoming on Passes, and decided ones, with what a trial recorded, under Past.
Related
- Predict satellite pass data volume: how
pass-yieldprices a pass - Rain fade at ground stations: what
rain-fade-swapreacts to - Audit passes against forecasts: what
audit-ledgercompares - MODCOD planning for VCM and ACM: what
modcod-planschedules
Next: load the demo fleet and test pass-yield on it, then make it your own in the builder.