eng-0123
window recorded
citable URL: https://halobench.com/records/eng-0123/ — this address never moves; the anchor /records/#eng-0123 keeps resolving
0.80 Whmean 137.51 W − 10.1 W idle floor (box-idle-all-empty) = 127.41 W attributable
window 2026-08-16T09:04:48Z → 2026-08-16T09:05:09Z · 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
covers run-0328
aihydra laguna-s21-fullbench house 4-item guard against cfg-0092 (the tau2 serving config). 21s window per run-meta.jsonl's own "guard-tau2-config-rocm-c32768" entry (09:04:48Z-09:05:09Z; the preceding 72s server boot is not in this window). Counter interpolated from raw ~10s-resolution HA history samples pulled fresh this session: start 21.80933683257663 kWh, end 21.810138981607825 kWh -- delta 0.802149 kWh*1e-3 = 0.8021 Wh. DISCREPANCY FLAGGED: MODEL-PAGE.md's draft reports this window as 0.84 Wh / 143.8 W. Recomputing the same window (09:04:48Z-09:05:09Z) directly from the raw ~10s HA samples straddling it (09:04:39.9Z/49.9Z/59.9Z/09:05:09.9Z local) gives 0.80 Wh / 137.5 W instead -- a ~5% difference on this one short window only. Every other window in this job's energy join (matrix cells, the 26-task tau2 run itself) reproduced the draft to within rounding; this guard window is the one exception found. Recorded per house policy: recompute from raw data rather than copy the delegate's number: this record uses the recomputed 0.80 Wh / 137.5 W figure, not the draft's 0.84 / 143.8.
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything