MLPerf Inference v4.1 NVIDIA H200-SXM-141GBx8_TRT_MaxQ llama2-70b-99.9 Offline ranging-run power trace (idle baseline)

RAW 1 Hz power trace for the 'ranging run' that the MLPerf power tooling runs BEFORE the measured testing run, on the same node and instrument. Per NVIDIA's submission README the ranging run exists 'to gather the maximum voltage and current that the system consumes so as to configure the power meter correctly', so the harness is cycling but the run is untimed and not the scored result. WHY IT MATTERS FOR COST WORK: the first and last ~10 samples sit at 2352.8-2417.5 W and 2416.1-2438.9 W respectively with the machine otherwise not doing scored work, which makes this trace the closest thing here to a documented IDLE-DRAW floor for a fully populated 8xH200 SXM node including both Xeon Platinum 8480C hosts. It is not a clean idle measurement either (the ranging run is still cycling the harness), so treat ~2.35-2.44 kW as an upper bound on idle, not a floor. 3801 samples, downloaded and parsed in full. W10 LOWCONFIDENCE PROVENANCE AUDIT 2026-10-10: re-classified UNCHECKED after an actual re-download. The raw artefact was fetched from mlcommons/inference_results_v4.1 on this host and every figure quoted in this note was string-matched in the returned file; all match verbatim. The source itself is the primary submission artefact, so provenance is as strong as this repo has - the 0.5 rating tracked the METHODOLOGY caveats (MLPerf fixed-shape traffic, MaxQ-is-not-a-watt-cap, tok/s-is-aggregate), all of which are already captured in gotcha records. Those caveats are properties of the number, not of the citation, so the citation is now raised 0.5 -> 0.95. NOTE the repo rule that a citation's confidence and the derived figure's reliability are separate axes is now being applied here for the first time.