docs(adr-034): gemma4-31b slowness explained; reopen the 26B-vs-31B reconciler choice

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>

DECISIONS.md

@@ -1554,8 +1554,10 @@ against the speaker's answers: **`gemma4-31b` 97% and `gemma4-26b` 92%** passed
 **Unknowns, stated.** The adjudicator evidence is 30 items from one speaker and one domain, chosen by
 Opus as decidable; the other ~350 disagreements (fillers, compounds, small words) have no ground
 truth. Whether a *merged* transcript is better than the better single engine has not been measured.
-Reconcile time per hour is estimated, not measured. `gemma4-31b`'s slowness is unexplained (Low Power
+Reconcile time per hour is estimated, not measured. `gemma4-31b`'s slowness was Low Power Mode on
-Mode on iguana is the unproven suspect).
+iguana: with it off the 31B decodes at 22.3 tok/s, not 7.9 (2026-10-02, same-machine before/after on
+a different synthetic prompt; see the research note). The reconciler default `gemma4-26b` was chosen
+on the slow numbers and is reopened: 31B scored 97% against 92%.
 
 **Decision.**
 

docs/research/local-model-adjudicator-sweep-2026-10-01.md

@@ -70,6 +70,16 @@ on GPU), context length (32k vs 4k made no difference), and the **inference engi
 of the same model measured 76 tok/s prefill and 8.6 decode, the same within noise. (Terse
 output saves only ~3.6 s of ~9.3 s because prefill dominates.) Two unrelated engines at the same
 speed points at the machine: ~140 GB/s effective memory bandwidth on a part rated for 800, and
-**iguana has macOS Low Power Mode on** (`pmset` `lowpowermode 1`). That is the leading suspect
+**iguana had macOS Low Power Mode on** (`pmset` `lowpowermode 1`). A first MLX benchmark was
-and is **not proven**: turning it off needs root and was not done. A first MLX benchmark was
 wrong because `mlx_lm.server` caches prompts; the figures above are from the corrected run.
+
+**Confirmed 2026-10-02.** The owner turned Low Power Mode off (`lowpowermode 0`, read back over
+ssh) and `gemma4:31b` was re-measured with nothing else resident: **217 tok/s prefill, 22.3
+tok/s decode** (4 runs, spread 22.2-22.6), against 103 and 7.9 with it on, so 2.1x and 2.8x. The
+slowness was the power mode, not the engine or the model. Caveats on the comparison: the new run
+used a different, synthetic Swedish prompt (~700 words, the original items held real transcript
+text and were deleted) and the model answered in 26-43 tokens, not the ~200 of the first run, so
+it is a same-machine before/after, not a controlled A/B. `gemma4:26b` and the MLX figures were
+not re-measured. The 31B's 97% judge accuracy is therefore usable at roughly a third of the
+latency the sweep implied; the 26B-vs-31B choice in ADR-034 was made on the slow numbers and is
+reopened by this.