Got it — sounds like you’re seeing “Late” flow runs for the past few days, but when you click the flow run IDs there’s nothing in the UI. A few things to know and a quick path to stop the noise: What “Late” means (Prefect 3.x) - “Late” = the scheduled start time passed (default >15s) and the run hasn’t been picked up by a worker yet. It’s still a Scheduled run, just flagged as late. - Default threshold = 15 seconds. - Common causes: no/overloaded workers, worker polling the wrong queue, concurrency limits, paused pool/queue. Why you might see “no data” for those IDs - The run may have transitioned out of Late quickly (to Pending/Running) and you’re viewing a different filter in the UI. - The link/ID may point to a different workspace or the run was deleted. - If you’re using notifications/automations, the message may arrive after the state already changed. Fast ways to reduce or stop “false” Late flags - If you’re self-hosting Prefect Server: - Increase the Late threshold:

export PREFECT_API_SERVICES_LATE_RUNS_AFTER_SECONDS=60
    # Restart your server so the setting takes effect
    prefect server start

- Or disable the Late service entirely:

prefect server start --no-late-runs

- If you’re on Prefect Cloud: - You can’t change the server-level Late threshold, but you can reduce noise with Automations: - Trigger: Flow run state change → Late - Hold “for” 1–2 minutes before notifying so brief delays don’t alert - Or auto-cancel after N minutes if that’s preferable - You can also adjust notification rules to ignore Late and only alert on Failed/Crash/etc. Quick checks to confirm what’s actually Late - List late runs:

prefect flow-run ls --state LATE --limit 50

- Inspect a specific run:

prefect flow-run inspect <run_id>
  prefect flow-run logs <run_id>

- Check work pools/queues and concurrency:

prefect work-pool ls
  prefect work-pool inspect <pool_name>

Review worker logs around the scheduled time to confirm the worker is polling the correct queue and not hitting concurrency limits. Helpful docs - Prefect docs - Automations (add a “for” duration to Late triggers) - States (Late vs Scheduled vs Pending) To help pinpoint the root cause and stop the flags for good, could you share: - Are you on Prefect Cloud or self-hosted Server? - A couple example flow run IDs and the workspace name/slug where you’re clicking the links - Your worker/agent Prefect version (

prefect version

) - The work pool and queue names those runs are using I can then tell you exactly why those are being marked Late and propose the minimal change to stop the false alerts.