Thanks for the analysis with examples here @Bananeweizen ! Agree that it looks off, and would be better if we ensure these numbers line up with the number of files, as opposed to number of files * number of recipes (?). I'll raise this to the team!

I see that only particular recipes produce the wrong sourceFiles stats

In the test repo are 12056 files.
For org.openrewrite.staticanalysis.CommonStaticAnalysis, the stats were correct.
For org.openrewrite.java.migrate.UpgradeToJava17 it doesn't:

At a quick glance those all seem to be multiples of the number of files, where AddJaxwsRuntime likely only runs once, whereas the others are configurable and may thus be included more than once.

What are you proposing in terms of a fix?

Yes, it seems that if a recipe is run with different parameters, the number of files visited is summed up.

In terms of fix:

• trying to create a unit test to fail (because initial unit tests passed successfully)
• trying to find the place where the multiplication happens

The counter issue is related to recordEdit method:

When the same recipe is executed several times with different parameters, the tag("name", recipe.getName()) will be the same.
The recordEdit triggers CumulativeTimer#recordNonNegative method, which automatically increments the counter for the same recipe:

So, we can't simply change the CumulativeTimer#recordNonNegative behaviour.
If we skip recordEdit for the same recipe, I assume, this will also impact the time statistics.

Perhaps, there is some other suitable Timer (or we can implement it).
Another option is to create different tags for the same recipes with different parameters, but before flushing Rows, we should collapse and aggregate statistics for the same recipe.

// @timtebeek

Unit test with the reproducer https://github.com/openrewrite/rewrite/compare/stats-issue-4876

Thanks again for pointing out this mismatch @Bananeweizen !