eng-0188
window reconstructed
citable URL: https://halobench.com/records/eng-0188/ — this address never moves; the anchor /records/#eng-0188 keeps resolving
84.18 Whmean 99 W − 10.1 W idle floor (box-idle-all-empty) = 88.9 W attributable
window 2026-08-16T15:11:44Z → 2026-08-16T16:02:45Z · 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 recorded window) · no contention
covers run-0356
Aggregate failure-window energy for both allowed Vulkan d131072 attempts. Exact attempt boundaries came from the preserved matrix timestamps and run-meta records: attempt 1 15:11:44Z..15:35:36Z and attempt 2 15:35:36Z..16:02:45Z, both rc=134 and contention=false. The combined counter endpoints were linearly interpolated from raw samples to 22.343875097 and 22.428054755 kWh. Both attempts ended in DeviceLostError and produced no throughput; this is failure-window Wh only, never throughput energy or Wh per correct answer.
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything