Explore workflow maps

Updated 4 hours ago • October 18, 2026

Workflows combines source-published structure with execution observations. Use it to understand the graph before a run and investigate how real traffic moves through it.

Declared and observed connections

Publish the catalog with your application release using Connect Your Project. Source discovery can show resolvable child calls before traffic exists. Unresolvable dynamic targets rely on runtime observations.

Observed calls overlays recorded child calls and their metrics. Disable it to concentrate on source-discovered structure; this does not remove source connections. Purple dotted child calls return to their caller; blue dashed handoffs transfer execution.

Good to know: A declared connection can appear before it has run. Publish topology to show paths from source; open a real run to see which path an execution took.

Inspect a selection

Select a workflow, node, or connection to open its inspector. Inspect the available version, calls, success/failure, latency, usage, and cost. A connection with recorded traffic can link to an example execution. An unexecuted connection shows source/target details.

Metrics and scope

Environment, time range, workflow/version, and metric selection change the displayed observations. Refresh metrics after publishing a catalog or generating new traffic. Missing metrics mean no matching evidence is available; they do not prove zero latency or cost.

Success/failure describes recorded execution outcomes. Tool denials and handled faults should be investigated through the run trace before treating them as workflow failures. Compare metrics over equivalent filters and versions.

Workflow health

Workflow health summarizes recorded activity for the selected period. Unknown means no matching runs. Healthy means an error rate below 5% and an interrupt rate below 25%. Degraded means an error rate of at least 5% or an interrupt rate of at least 25%. Failing means an error rate of at least 20%. Idle means the latest recorded activity is more than seven days old.

The Health view's emphasis controls highlight individual nodes using separate thresholds: error rate above 3%, p95 latency above two seconds, or interrupt rate above 10%. Volume emphasizes recorded run counts; cost shows recorded average cost. These highlights do not redefine workflow health.

Node states

Hover or focus a state symbol to read its explanation. An interrupted node's pause symbol means at least one run paused for input at that node. A retry uses an orange dot, border, and circular arrow. A failure means at least one run containing the node failed; it does not by itself identify the failing operation. A warning represents a recorded warning signal.

Healthy means no failure, pause-at-this-node, or retry signal was recorded for matching runs. No node activity means the node did not run in the selected period. These are period summaries; open a run to inspect its event sequence. The run topology instead shows the node's state within that individual execution.

Next: Metrics, Runs, or Child workflow integration.