https://www.boost.org/doc/contributor-guide/testing/fuzzing.html describes the fuzzing process in great detail citing Boost.JSON CI configs as a good example

However it seems the persisting and restoring of the corpus is not working, is it?

• The cache action restores head/fuzzing/corpus.tar
• The pack step puts the cmin folder into corpus.tar.tmp, then moves it to head/fuzzing
• The upload-artifact step uploads head/fuzzing/corpus.tar

The crucial missing steps are the extraction of a potentially restored corpus and using corpus.tar as the file name.

See this run

Warning: No files were found with the provided path: head/fuzzing/corpus.tar. No artifacts will be uploaded.