eng-0152
window recorded
citable URL: https://halobench.com/records/eng-0152/ — this address never moves; the anchor /records/#eng-0152 keeps resolving
59.67 Whmean 157.94 W − 10.1 W idle floor (box-idle-all-empty) = 147.84 W attributable
window 2026-08-17T11:46:56Z → 2026-08-17T12:09:36Z · provenance recorded
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 run-meta window) · no contention
Live join (protocol §11, joined same session as the run per the job brief's "join Part 1/2's own arms live as you go") for ornith-35b Q8_0 Vulkan d262144 model-max (cfg-0106), covering both the prefill row (run-0383) and decode row (run-0384). Window is run-meta-recorded (run_begin/run_end used directly this time, unlike the Part 3 backfill this session also joined), so window_provenance is "recorded" not "reconstructed". Counter interpolated from raw ~10s-resolution HA history samples: start 25.269026 kWh (11:46:56Z), end 25.328694 kWh (12:09:36Z). Lower mean draw than the ROCm arms in this session's eng-0147..eng-0150 (157.94 W vs ~173-179 W), consistent with eng-0151's same cross-backend pattern at d204800.
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
candidate gate history ornith-35b