The Behavior tab shows how each step of your workflow actually behaved as it ran across your work items: which tools each run used and how often, how consistent that behavior was, and which runs stood out. Use it to spot unusual activity and find opportunities to make a workflow more reliable.
This is different from reviewing how a workflow is set up — its instructions, tools, and steps. The Behavior tab looks at what actually happened when the workflow ran.
Where to find it
Open your thunk, go to Monitor, and select the Behavior tab. Monitor is available to thunk admins.
Running an analysis
The analysis does not update on its own. Click Run analysis at the bottom of the list on the left to analyze the thunk's recent activity. While it works the button reads Analyzing...; when it finishes, the new report appears at the top of the list and opens.
Your 10 most recent analyses are kept in that list, newest first, so you can go back to an earlier report and compare how a step behaved before and after you changed it.
How steps are grouped
Steps are shown in two sections:
Workflow steps — the steps that run once for each work item. This is the per-item work of your workflow, and it is where you will usually spend your attention.
System steps — setup and background steps (for example, preparing content folders or building a tool). These are not part of the per-item workflow, so they are listed separately.
Within each section, steps appear in the order the workflow runs them.
Reading a step's row
Each row summarizes one step across all of its runs.
Runs — how many times the step ran (once per work item).
Tool count — the fewest, typical (median), and most tools a single run used.
Tool Consistency — how often runs followed the same sequence of tools, shown as a percentage along with the number of distinct patterns seen. A high percentage means the step does the same thing nearly every time; a low percentage means it varies from run to run.
Looking at a single step
Select a step to expand it. By default you will see its tool frequency — for each tool the step used:
in how many runs it was called (so you can see whether a tool is used every time or only occasionally);
the average number of times it was called per run;
the total number of calls across all runs; and
the share of its calls that returned an error, when the report has that information (hover over it to see the number of calls).
This shows at a glance which tools a step relies on, and whether any tool is being called far more often than you would expect.
For a closer look, open Advanced: tool-call sequences. This shows the exact order of tools each run used — the most common sequence, and the full distribution of every distinct sequence with how many runs followed it. Sequences that stood out are marked, so you can see exactly which ones returned an error, used an unusual number of tools (an outlier), or were a one-off that appeared in no other run.
Putting it to use
Consistent steps are behaving predictably. A step that should be routine but shows low consistency is worth a closer look — it is doing something different each time.
Signals point you to specific runs. Errors, outliers, and one-off sequences highlight the runs that did not go as expected, so you can investigate those rather than reviewing every run.
Compare actual tool use to what you intended. If a step is reaching for tools you did not expect, or repeating tools more than it needs to, that is a cue to refine the step's instructions or the tools available to it. See Tuning for reliability for how to make those adjustments.
What the analysis covers
The analysis looks at runs that have settled — finished, or paused waiting on a person — and leaves out runs that are still actively in progress, so their partial activity does not skew the numbers.
For workflows with a large amount of activity, it focuses on the most recent runs of each step rather than the entire history, and shows a note when it is doing so. Every run it shows is complete; you are seeing a recent sample, not a partial one.