Predictions in a pipeline
Queue a forecast from a pipeline without confusing its ingress with model evidence
A Data Pipeline Prediction node queues a project forecaster. The forecaster still loads its own pinned evidence query; an incoming source edge does not replace that query or turn every row into a forecast.
Equipment scenario
You need workspace source access, pipeline authoring/run permissions, a project with an active equipment forecaster and a published evidence query accepting P-101. Forecasting/model/workflow execution must be configured. Prepare an equipment source endpoint and inspect P-101's actual records; credentials stay in the plugin installation.
Create an Ingress → Prediction path in Studio. Use the structured source's Full load/parser ingress with a declared equipment collection output connected to the Prediction input. The graph needs a valid ingress and connected execution path.
Configure the prediction node
Select the existing Equipment failure risk Forecaster. Set Subject P-101, Horizon 7d, a future evaluation date seven days after the intended run and empty params. Inspect the forecaster's evidence pin. Validate, save a graph version and run once.
Ask the Studio assistant to add a prediction for existing Equipment failure risk, literal subject P-101, horizon 7d and an explicit matching future horizon end. Require its published evidence query; do not substitute incoming rows as evidence. Review the node and graph validation before saving/running.
The predict section belongs inside a complete Studio graph document. Replace the forecaster UUID and illustrative date. It is not a standalone MCP/API request.
{"forecasterId":"<EQUIPMENT_FORECASTER_UUID>","subjectRef":"P-101","horizon":"7d","horizonEndsAt":"2027-01-08T08:00:00Z","params":{}}Inspect both runs
The node returns predictionId and nonBlocking true. Pipeline success means the invocation was queued; the pipeline does not wait for final forecast output. Open the prediction and inspect its own Completed/Failed/Cancelled state, evidence, output and evaluation date.
Current v1 supports a literal configured subject, not a field-derived subject/per-row forecast loop. Use watchlists or explicit individual runs for repeated equipment subjects. Re-running a graph with a fixed horizonEndsAt can eventually fail once the date is in the past; review the date for each intended window.
The Data Pipeline prediction guide explains the surrounding graph lifecycle. This page states the equipment-specific prerequisite and node behavior without claiming that the source edge supplies sufficient predictive evidence.