eng-0147
window reconstructed
citable URL: https://halobench.com/records/eng-0147/ — this address never moves; the anchor /records/#eng-0147 keeps resolving
108.4 Whmean 173.64 W − 10.1 W idle floor (box-idle-all-empty) = 163.54 W attributable
window 2026-08-17T07:15:40Z → 2026-08-17T07:53:07Z · provenance reconstructed
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 session window) · no contention
Deadline-critical batch join (protocol §11) for the 2026-08-17 depth-ladder backfill, qwen38-27b Q8_0 ROCm d204800 (cfg-0103), covering both the prefill row (run-0373) and decode row (run-0374) from the single llama-bench invocation. Session window reconstructed from ~/bench-results/depth-backfill-20260817/qwen38-27b/d204800.started (07:15:40Z) and the output JSON's file mtime (07:53:07Z) -- the backfill wrapper this session used did not call run_begin/run_end for every cell (only its first cell, qwen36-27b-mtp, has a run-meta.jsonl entry), so window_provenance is reconstructed, not recorded. Counter interpolated from raw ~10s-resolution HA history samples: start 24.838408..24.838411 kWh (07:15:40Z), end 24.946790 kWh (07:53:07Z).
Wh, never joules — read as a cumulative-counter difference across the window; delta over the idle baseline is the only figure that means anything