Home › Evidence › Records › eng-0107
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

eng-0107

window recorded
citable URL: https://halobench.com/records/eng-0107/ — this address never moves; the anchor /records/#eng-0107 keeps resolving
534.3 Whmean 171.9 W − 10.1 W idle floor (box-idle-all-empty) = 161.8 W attributable
window 2026-08-15T22:06:31Z → 2026-08-16T01:13:01Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy, 5-minute statistics interpolated to the run-meta window) · no contention
per task 24.2853 Wh · 0.7358 p at 30.3 p/kWh
covers run-0287
aihydra run-meta.jsonl entry deepseek-v4-flash-fullbench/tau2-full-26task (rc=0 — the run completed all 26 requested tasks cleanly, NOT a wall-bound cut; contrast qwen38-27b's eng-0081, which was cut at rc=124). wh_per_task is Wh-per-CORRECT-ANSWER: wh_total / tasks_passed (22 of 26, run-0287.metrics). This is WHOLE-SESSION energy, not agent-only — the window includes simulator (user_llm) wait time end to end. Joined same-session (not retroactive): HA statistics pulled directly against the exact run-meta window immediately after the run completed.

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

model pages deepseek-v4-flash