This is a demo with invented workflows. Install GreenLight to see your own.

• Silent error

  raises critical1 finding in this scan

  applies to Active workflows with a recent successful execution

  Question

  Did a step fail inside a run that reported success?

  How it decides

  Opens a sample of the most recent successful executions and reads the output of every node. A node that emits an error as ordinary data is reported, even though the run finished green. This usually happens when a step has continue on fail enabled. A workflow that is switched off is not checked, because its history may hold old failures nobody needs to act on.

  sample

  5 successful executions by default

  history needed

  none

  What to review

  Open the named node and read the error it returns. Check the credential or service behind it, then decide whether the workflow should stop on that failure instead of carrying on.

• Duration drift

  raises warning1 finding in this scan

  applies to Any workflow with enough history

  Question

  Is this workflow suddenly much slower than it used to be?

  How it decides

  Compares the median run time of recent executions with the same workflow’s own history. Either condition alone reports ordinary variation, so both must hold.

  recent median

  above the historical 95th percentile

  recent median

  at least twice the historical median

  baseline runs

  10 or more

  recent runs

  3 or more

  What to review

  A step is probably retrying or waiting on a downstream service before it gives up. Compare a recent slow execution with an older fast one and look for the node whose time changed.

• Silence

  raises critical1 finding in this scan

  applies to Active workflows started by a clock

  Question

  Has a scheduled workflow stopped running without being switched off?

  How it decides

  Compares the time since the last run with the usual interval between runs. Workflows started by webhooks or polling triggers are not judged, because a quiet period can be normal for them.

  quiet for

  more than 4 times the usual interval

  runs needed

  3 or more, to learn the interval

  What to review

  Check that the workflow is still active and that its schedule trigger is still registered. Then look at the instance logs around the time of the last run.

• Frequency drop

  raises warning1 finding in this scan

  applies to Active workflows started by a clock

  Question

  Is it still running, but far less often than before?

  How it decides

  Counts the runs in the recent window and compares them with what the workflow’s own history predicts. A workflow that did not run at all in the window is judged by the silence check instead, which compares the gap against the workflow’s usual interval.

  window

  24 hours

  reports when

  fewer than half the predicted runs

  baseline runs

  10 or more

  What to review

  Compare the schedule trigger as it is configured today with the cadence the history shows. An edited interval is the usual cause.