Results out: your MCS and destinations
Results leave Constellation two ways. Your mission control system (MCS) pulls them: a connector inside your network reads your account's workflow results, outbound, and writes into your system. Destinations have them pushed: Constellation sends results to your Amazon S3 bucket, Grafana Cloud stack or webhook. Whichever you use, every result is also in the API, read with GET /predictions.
| Your MCS, through a connector | A destination | |
|---|---|---|
| Systems | Yamcs, Open MCT, OpenC3 COSMOS, Anduril Lattice, your own | Amazon S3, Grafana Cloud, webhooks |
| Who connects | The bridge in your network calls Constellation over HTTPS; nothing connects in | Constellation calls your bucket, stack or endpoint |
| On a workflow | A mission control output (feed in the API) naming the bridge's connection | A destination output naming the destination |
| Credentials | Your system's stay with the bridge, which holds a key with predictions:read (Lattice also config:read); the Open MCT plugin reading in a browser holds a key with only results:read | An IAM role your account trusts, a Grafana token, or a signing secret |
| Recommendations | Advice, or an instruction once someone approves it (approvals need Operator + Act) | Advice; a recommendation held for approval arrives as pending and is not sent again |
| Per account | 50 connections | 3 destinations |
| Plan to deliver | Operator; Lattice and COSMOS on Enterprise | Operator |
Test runs and scenario runs deliver nothing either way; their results stay in the API.
Your MCS pulls workflow results
- A workflow's mission control output names a connection, such as
yamcs-bridge, and how that system should show results: event wording, timeline bands, parameters. - When a run finishes, its results for that connection join your account's workflow results.
- The bridge for that connection, in your network, pulls them and writes into your system. Results are keyed by pass, so a newer run updates what an earlier one wrote.
- The bridge reports what it wrote, and Integrations shows each connection's last read and check-in.
On Integrations, Create its key on a system makes its bridge's key with these scopes. For Yamcs, COSMOS and Lattice the system then joins the Flow under Results out as Not reading yet until its bridge first reads.
Workflow results keep each run for 7 days. Connectors write forecasts as forecasts, never over your own schedule or telemetry. Install steps, versions and the rules each connector keeps are in Connectors; building your own is in Your own connector and the results protocol.
Destinations
A destination receives two kinds of results:
- Workflow runs whose graph has a destination output naming it, from each run that delivers.
- Served forecasts: the latest link SNR forecasts a live key of a self-serve account reads by
link_idsonGET /predictions, directly or through an agent over MCP. Reads by instant (at),as_of_utcreads, test-key reads, enterprise accounts' reads and the shared demo account's reads are not sent.
Deliveries are not counted as predictions. An account has at most 3 destinations, of any type; a fourth is refused with 409 destination_limit_reached. A webhook needs Operator to create, and a workflow's destination output needs Operator to deliver.
Each delivery carries an id that stays the same if it is delivered again: an S3 object is overwritten under the same key, a Grafana point lands on the same sample, and a webhook carries it as Idempotency-Key. See Webhooks and SDKs.
Choose
- Operators should see results where they already work: use a connector.
- Analysts, dashboards or your own services: use S3, Grafana Cloud or a webhook.
- Code that asks when it needs to: read the API,
GET /predictions?workflow=<slug>, which needs no output at all.
Next: add an output on the builder, or in the graph's JSON as in the node reference.