eng-0280
window recorded
citable URL: https://halobench.com/records/eng-0280/ — this address never moves; the anchor /records/#eng-0280 keeps resolving
205.2 Whmean 186.56 W − 10.1 W idle floor (box-idle-all-empty) = 176.46 W attributable
window 2026-09-06T09:27:14Z → 2026-09-06T10:33:13Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy) · no contention
per task 10.258 Wh · 0.31 p at 30.3 p/kWh
covers run-0665
The pwilkin candidate's τ² 0-25 window (cfg-0185, run-0665). Counter differenced on the HA energy sensor: 48.72781343 kWh at 09:27:14Z -> 48.93297700 kWh at 10:33:13Z = 205.16 Wh over 1.10 hours, mean 186.6 W. wh_per_task is Wh-per-CORRECT-ANSWER: wh_total / tasks_passed (20 of 24 scored, run-0665), the same method as eng-0081 / eng-0278 so the three compare directly: 10.26 Wh per correct answer, 0.31 pence at 30.3 p/kWh.
Well below the worker's 15.18 (eng-0278) for two reasons, neither a clean A/B: the pwilkin stack decodes faster (66-min window vs the worker's 106 min -> less total energy) and got fewer answers right (20 vs 22). Whole-session energy (includes user-sim wait). Baseline is the canonical 10.1 W aihydra floor (see eng-0278's baseline caveat).
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-0134
model pages qwen38-27b