Different dependency graph between tests

I’m running 2 different tests, and the second one fails because the dependency graph is incomplete (lib_lbf_mode = chain+ seems to be ignored) but the first one succeeds and the graph is complete. I’m running out of ideas why might this be, shouldn’t the graphs be identical? The tests used to work until I tried splitting them into 2 different tests.

For reference Firebase library add issue arduino mobitz is the exact issue I’m having. The suggested fix worked until I tried splitting the test from one file to two separate files making 2 tests which build separately.