From #29951 (comment) (@ogrisel)

Somehow related, side note: maybe check_estimator could be made to return a structured result object that is programmatically introspectable.

This would allow third-party libraries to build and publish a scikit-learn compliance table to be integrated as part of their documentation. In case of XFAILed checks, the reason could be displayed as part of the report.

Currently, users would have to dig into CI logs and grep the pytest output, assuming those projects use the parametrize_with_checks as part of a pytest test suite instead of just calling check_estimator.

Thinking about, we could even start by eating our own dog food: we have no simple summary of all the XFAILed/XPASSed cases for scikit-learn estimators at the moment.

I like the idea, but in order not to forget about it, creating its own issue.