Two measurements that reported on something they never varied

Two independent measurements on the same day produced confident results without ever changing the thing they claimed to be measuring. A mechanised identity test deleted each module and recorded whether the program survived — but every module the entry point imports raises ImportError when deleted, so the test measured reachability and reported it as structural necessity; the two modules it did flag were flagged because they were not wired in at all. Separately, a comparison of CPython json with and without its C accelerator reported the two implementations byte-identical while both runs used the C implementation, because json/scanner.py binds `make_scanner = c_make_scanner or py_make_scanner` at import time and clearing c_make_scanner afterwards changes nothing.