eng-0176
window reconstructed
citable URL: https://halobench.com/records/eng-0176/ — this address never moves; the anchor /records/#eng-0176 keeps resolving
876.5 Whmean 166.9 W − 10.1 W idle floor (box-idle-all-empty) = 156.8 W attributable
window 2026-08-10T09:07:30Z → 2026-08-10T14:22:34Z · provenance reconstructed
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy, raw ~10s-resolution history samples, linearly interpolated to the transcript-derived window) · no contention
per task 87.6496 Wh · 2.6558 p at 30.3 p/kWh
covers run-0102
aihydra nemotron3-super UD-IQ4_XS tau2-bench-airline 16-scored-task arm window (2026-08-10T09:07:30Z..2026-08-10T14:22:34Z, 18904s), taken from run-0102's started_at/ended_at. run-0102 predates matching run-meta instrumentation for this arm, so the window is reconstructed from transcript timestamps rather than a recorded meter window. Counter interpolated from raw ~10s-resolution HA history samples pulled fresh 2026-08-18 as part of the run-0101/run-0102/run-0111 batch: start 6.663011680 kWh, end 7.539507967 kWh. The transcript contains internal idle gaps between task groups and one unscored infrastructure-error simulation; the gaps are included end-to-end as wall-clock session energy. wh_per_task is Wh-per-CORRECT-ANSWER: wh_total / tasks_passed (10 of 16 scored tasks, run-0102.metrics) = 87.6496 Wh, 2.6558 pence/correct at the standing 30.3 p/kWh tariff. This arm ran an unpinned self-play user simulator (user_llm is the same nemotron3-super path as the agent), so it remains provisional per clm-0043.
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything
Cited by — computed at build time, never stored
claims clm-0042