eng-0230
window recorded
citable URL: https://halobench.com/records/eng-0230/ — this address never moves; the anchor /records/#eng-0230 keeps resolving
83.39 Whmean 137.7 W − 10.1 W idle floor (box-idle-all-empty) = 127.6 W attributable
window 2026-08-22T20:57:25Z → 2026-08-22T21:34:00Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy, raw ~10s samples linearly interpolated to the recorded window) · peak 215.36 W · no contention
per task 3.2073 Wh · 0.4211 p at 30.3 p/kWh
covers run-0569
HO-009-AB PLAIN arm tau2 full-26 (reasoning OFF). Counter interpolated from 33.100245 to 33.183670 kWh across the plain tau2 window 20:57:25-21:34:00Z on 2026-08-22. 6/26 tasks passed, so wh_per_task uses 26 and pence_per_correct_answer uses 6 correct tasks as the denominator. Power profile shows a single isolated active band (idle ~11.5W before 20:55Z and after 21:34Z) — no contention with the prior LLaDA external job, which had already released. Build 7077abb, ROCm0/gfx1151, f16/f16 KV, -rea off. Plain-vs-n2 capability comparison only; no production or throughput claim.
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything