eng-0082
window recorded
citable URL: https://halobench.com/records/eng-0082/ — this address never moves; the anchor /records/#eng-0082 keeps resolving
163.7 Whmean 137 W − 10.1 W idle floor (box-idle-all-empty) = 126.9 W attributable
window 2026-08-15T05:22:15Z → 2026-08-15T06:33:58Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy) · no contention
per task 14.8814 Wh · 0.4509 p at 30.3 p/kWh
covers run-0272
aihydra run-meta.jsonl entry kv-quality-patched/f16 (2026-08-15T05:22:15Z, rc=124 — the hard-split design caps this arm at half the total time budget by construction, per bench/queue-kv-quality-patched.sh; the cut is the intended stopping point, not a failure). wh_per_task is Wh-per-CORRECT-ANSWER: wh_total / tasks_passed (11 of 14 scored, run-0272.metrics; 2 of the 16 attempted sims hit infrastructure_error and are excluded from tasks_total per run-0272). Paired against eng-0083 (q8_0, same 14 task-matched ids): see clm-0058 for the quality comparison and the Wh-per-correct reading together. Retroactive join computed 2026-08-15.
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything