Watchlists
Schedule named subjects and keep forecast windows separate from run cadence
A watchlist item schedules one named subject for a forecaster. Daily prediction cadence and a seven-day outcome window are different settings: the item may run every day while each prediction is evaluated seven days after its own invocation.
Equipment setup
You need a Semogram account with project access, an Active equipment forecaster and an evidence query that accepts P-101. Verify a single manual run first. The deployment must run the authorized prediction scheduler; saving a watchlist item alone does not start a scheduler.
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "forecast_watchlist_create",
"arguments": {
"forecasterId": "<EQUIPMENT_FORECASTER_UUID>",
"subjectRef": "P-101",
"horizon": "7d",
"params": {},
"schedule": "daily",
"evaluationHorizon": "7 days",
"enabled": true,
"idempotencyKey": "<UNIQUE_WATCHLIST_KEY>"
}
}
}Open the equipment forecaster's Watchlist page. Add subject P-101, schedule daily, forecast horizon 7d and evaluation horizon 7 days. Inspect enabled state and next run. Use the watchlist's edit controls to pause or adjust it. Parameter customization, when needed beyond the exposed controls, is available through MCP.
Ask a connected assistant to prepare a daily watchlist item for Equipment failure risk, subject P-101, forecast horizon 7d, evaluation horizon 7 days and empty params. Review the forecaster reference and timing before it calls forecast_watchlist_create. Ask it to read back forecast_watchlist_list after saving.
Call forecast_watchlist_create with the arguments shown above. Scope comes from the project connection; the forecaster must belong to it. A successful response is an item, not proof that scheduled prediction work has run.
Current schedule behavior
The scheduler's supported interval labels are hourly, daily and weekly. They schedule from the tick's execution time, not a promised wall-clock appointment. The current database function treats other strings as a daily fallback; do not use arbitrary cron expressions expecting them to be parsed. Use these established labels and verify nextRunAt.
Evaluation horizon is an interval such as 7 days and is required for scheduling. Each run's horizonEndsAt is invocation time plus that interval. Forecast horizon is the separate prompt label. Keep those values aligned with your event definition.
A null nextRunAt makes an enabled item due on the next authorized scheduler tick. Scheduler timing, bounded batches and workflow dispatch affect when the prediction begins. Inspect actual lastRunAt, nextRunAt, prediction status and watchlistId.
Maintain and pause
Use forecast_watchlist_update with forecasterId, watchlistId, changed fields and a unique idempotencyKey. Set enabled false to pause. New scheduled runs use the forecaster's then-current executable version; older predictions keep their version snapshots. Disabling/archiving the forecaster prevents new scheduled selection but does not settle existing outcomes.
Removing an item uses forecast_watchlist_remove and is a destructive management action. Existing prediction records and their evaluations should still be inspected independently.
Operate the schedule
Scheduling predictions covers the complete UI/assistant/MCP setup, deployment checks, pausing and version changes. Follow Monitoring to confirm the resulting prediction and Outcome operations to check its eventual observation.