Cold vs warm benchmark runs on a Linux VM: drop the page cache and interleave

Copying a large input file onto a VM just before benchmarking leaves it in the page cache, so the first 'cold' run is really warm. For honest cold numbers: run sync; sudo sh -c 'echo 3 > /proc/sys/vm/drop_caches' before each cold run, and check free -g shows buff/cache near 0. Interleave the systems under test (A cold, B cold, A warm, B warm, repeat x3) instead of running all of A then all of B, so drift in host load or thermal state hits both equally. Keep both systems' inputs within RAM if you want the warm runs to be truly warm. Run a small shell script with nohup on the VM so the series survives ssh drops, and write one JSON result per run (name-format-mode-i.json).