docs: Jev candidate bench vs SemIf (fv-ml1 GPU 3) — Intern-Decision-4B is the replacement if SemIf is displaced

Operator ask relayed by brokkr-smithy-dev. Positive control (SemIf 187/231,
hard 0.613) reproduced exactly; negative control and a 4-restart noise floor
(0 flips) measured. On our replaced-baseline sets no candidate beats SemIf-with-
rotations beyond the ~4-pt floor; Intern-Decision-4B native matches it at one
ordering, is better on Wyrd, fits 9.7/10.3 GB and is 1.5-2.3x faster. JevBench
rank does not transfer. Raw per-item data kept out of git.

services/semif-serve/bench-jev-2026-09-30/code/Dockerfile.native

@@ -0,0 +1,10 @@
+# Transient bench image (2026-09-30): semif-serve:0.1.4's exact torch 2.10.0+cu128 / transformers 5.17.0 /
+# fla 0.5.2 / causal-conv1d 1.7.0 stack, plus what the candidates' own runtimes need. Removed after the bench.
+FROM semif-serve:0.1.4
+USER root
+RUN uv pip install --python /app/.venv/bin/python --no-cache --no-deps \
+      "jevk5 @ git+https://github.com/allebee/jevk5@85238d7be5527370c43206fe54cd752eb3134c1b" \
+ && uv pip install --python /app/.venv/bin/python --no-cache peft pillow python-multipart \
+ && uv pip install --python /app/.venv/bin/python --no-cache --no-deps \
+      --index-url https://download.pytorch.org/whl/cu128 "torchvision==0.25.0"
+USER semif