fix: prioritize preferred GPU routes with fallback

README.md

@@ -444,11 +444,18 @@ durable Git checkpoint per roughly 2,000 scanned rows. Raw rows are discarded
 after every aggregate merge. Transient API failure pauses at the current cursor
 for the next poll cycle; successful completion disables the scan permanently.
 
+Version `2026.09.18.1` keeps the first six ranked GPUs as the preferred routes
+for each model and only falls back to lower-ranked eligible GPUs when all six
+preferred routes are unavailable. Candidate generation retains at most six
+routes per model, preserving successful-route and queue ordering without
+discarding capacity on lower-ranked GPUs when the preferred tranche cannot run
+the model.
+
 ## Deploy
 
 Create a tag and submit the repository URL plus tag in "我的适配智能体".
 
 ```bash
-git tag -a agent-v33 -m "ModelHub agent 2026.09.04.4"
-git push origin main agent-v33
+git tag -a agent-v34 -m "ModelHub agent 2026.09.18.1"
+git push origin main agent-v34
 ```