Dispatch & Sprints
Choose the right Pulse AI dispatch path and protect scope, dependencies, and review movement.
Operating model outline
| Page | Reader job | Unique section | Media need |
|---|---|---|---|
| Workday | Run the active operating session | What the pool is allowed to do | Workday screenshot |
| Dispatch | Send board work into execution | Pool dispatch vs direct dispatch | Dispatch workflow diagram |
| Board | Read what is eligible | Dependency and blocker rules | Status flow visual |
| Review Queue | Decide what should run next | Review changes the next dispatch | Review checklist |
Before you dispatch
Dispatch starts from board truth. Confirm the task IDs, sprint, sprint group, dependencies, and current statuses before sending work to agents. If the visible request is a focused closeout, dispatch that slice. If the request is a scheduled workday, load the workday pool. If the request is one immediate action, use direct dispatch.
The safest question is: "What exact work should be allowed to start now?"
Pool dispatch vs direct dispatch
Use pool dispatch for scheduled workdays, planned sprints, and sprint groups that should drain through the Workday queue. Pool dispatch keeps a runtime pool, applies slot limits, and can subscribe to sprint mutations.
Use direct dispatch for immediate one-off work, ad hoc review, or a specific rerun. Direct dispatch starts the selected item without treating it as a whole-day agenda.
Do not self-throttle by dispatching only the number of currently open slots. The dispatcher owns slot limits. Your job is to send the approved scope once.
Duplicate and blocked-work safeguards
Dispatch should not create duplicate work just because a task looks stale. First check whether an existing owner task, review finding, dependency, or failed attempt already explains the state.
Blocked work should remain blocked until the upstream dependency is complete. Review-pending work should not be called unfinished; it is waiting for approval or a revision decision. Done work should not relaunch unless the task has been reopened with a concrete reason.
What to inspect when dispatch looks wrong
Check these in order:
- The requested task, sprint, or sprint group file.
- The first queued item in the pool.
- Dependency fields and blocked status.
- Review notes on the item that just completed.
- Dispatch logs or attempt metadata if duplicate terminals appear.
If the pool contains unrelated work, stop the Workday and repool the approved slice. If a task cannot be found but the file exists, reload or inspect the board cache before creating a duplicate.