Home › Evidence › Records › eng-0095
⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.

eng-0095

window recorded
citable URL: https://halobench.com/records/eng-0095/ — this address never moves; the anchor /records/#eng-0095 keeps resolving
5.25 Whmean 176.8 W − 10.1 W idle floor (box-idle-all-empty) = 166.7 W attributable
window 2026-08-15T13:45:53Z → 2026-08-15T13:47:40Z · provenance recorded
method wall-meter · aihydra energy sensor via Home Assistant (counter-difference on sensor.hardware_ai_hydra_energy) · no contention
covers run-0285
EOS-cliff cause-isolation matrix, avoidance probe 3 / cache_prompt:true on the failing baseline config (run-0285, cfg-0075, 15 requests, 12 degenerate — WORSE than the same config's own 11/15 without cache_prompt, run-0277); wh_per_task/pence_per_correct_answer left null (guard run). The short 107 s window reflects how quickly this cell collapses into 1-token generations. Counter samples: start 19.5945554226636776 kWh (shared boundary with eng-0094's end), end 19.5998108834028176 kWh (14:47:39 local, 1s before window close — nearest recorded tick). 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