`test_bazel_run_resmoke.py` invoked `bazel run` with `-config=local`, which sets `//bazel/config:use_rbe_sysroot=True`. `test_bazel_burn_in_end2end.py`'s bazel build uses `config=no-remote-exec`, which leaves that flag at its default `False`. Since the two tests can run back-to-back in the same suite, the flag flip forced Bazel to discard its analysis cache and reconfigure the whole target graph before the burn-in build, contributing to observed timeouts. Switching to `-config=no-remote-exec` (matching burn-in) keeps the flag stable across both tests.

*Evidence* — running `test_bazel_run_resmoke.py` then `test_bazel_burn_in_end2end.py` back to back, warm cache, no `bazel clean`, two runs each:

The savings are concentrated almost entirely in the burn-in build step (~10.6s → ~2s), consistent with the flag-flip/reconfigure theory above.