⚠ This notebook has stopped taking notes: newest record is 23 days old (eng-0280), against an expected cadence of 14 days.
clm-0116
measured-heremed ●●○
citable URL: https://halobench.com/records/clm-0116/ — this address never moves; the anchor /records/#clm-0116 keeps resolving
HK-001-FULLBENCH-REVIEW record for the affected v1.1 Kairic Edge TheRock configuration (cfg-0164): Qwen3.8-27B-IU4-Kairic-Edge.gguf on the TheRock 7.14 / AMD clang 23.0.0 ROCm runtime, c32768, --kairic-edge --no-mmap. The record-backed guard exercised three probes and passed them, with cross-request isolation N/A under --parallel 1 (run-0601).
Record-backed performance evidence retained here is the c32768 FIT (prefill 279.6 t/s, decode 11.73 t/s; run-0602) and the engine-path depth rungs at d0 (prefill 612.3 t/s, decode 13.14 t/s; run-0603) and d32768 (prefill 272.1 t/s, decode 11.66 t/s; run-0604). These are measurements of the affected v1.1 configuration and are not clean/v1.2 capability validation.
Served-path, deeper-depth, smoke, and full-26 attribution is on evidence hold in this correction. No replacement run IDs, metrics, or energy allocations are asserted.
HK-001-FULLBENCH-REVIEW retained only the locally resolved guard and performance records run-0601..run-0604. Other HK-001-FULLBENCH material is held by the correction packet; this record does not make a clean/v1.2 capability claim.
[2026-08-24 COMPARABILITY ANNOTATION — HK-V12-ANNOTATE, per t_b05473c9] The retained records use the affected v1.1 Kairic Edge runtime. The server_flags_degraded marker identifies that configuration state; it does not validate v1.2 capability or clear the evidence hold.