fix: prioritize preferred GPU routes with fallback

tests/test_submission_safety.py

@@ -68,6 +68,51 @@ class ExactGpuProcessedClient:
         return {"code": 0, "data": {"id": "should-not-submit"}}
 
 
+class AllGpuOfficialRegistry:
+    @staticmethod
+    def task_types_for(_client, *, model_address: str, model_last_modified: str | None, gpu: str) -> list[str]:
+        del model_address, model_last_modified, gpu
+        return ["text-generation"]
+
+
+class AllGpuVettedMarket:
+    @staticmethod
+    def has_framework_route(_task_type: str, _target_gpu: str) -> bool:
+        return True
+
+    @staticmethod
+    def ensure_framework_route(_client, _task_type: str, _target_gpu: str) -> None:
+        return None
+
+    @staticmethod
+    def selectable_frameworks(
+        *, task_type: str, target_gpu: str, incumbent_frameworks: list[str], inspection
+    ) -> list[str]:
+        del task_type, target_gpu, incumbent_frameworks, inspection
+        return ["vllm"]
+
+    @staticmethod
+    def official_config(
+        *, task_type: str, target_gpu: str, framework: str, gguf_filename: str | None = None
+    ) -> str:
+        del task_type, target_gpu, framework, gguf_filename
+        return "framework: vllm"
+
+    @staticmethod
+    def framework_metadata(_task_type: str, _target_gpu: str, _framework: str) -> dict:
+        return {"frameworkCombinedScore": 1.0, "frameworkOfficialConfigValid": True}
+
+    @staticmethod
+    def gpu_metadata(_target_gpu: str) -> dict:
+        return {}
+
+
+class SeventhGpuFallbackMarket(AllGpuVettedMarket):
+    @staticmethod
+    def has_framework_route(_task_type: str, target_gpu: str) -> bool:
+        return target_gpu == "gpu-7"
+
+
 class UniquenessRejectingClient:
     @staticmethod
     def add_task(_payload: dict) -> dict:
@@ -107,6 +152,50 @@ def candidate(target_gpu: str = "Vastai_va16") -> dict:
 
 
 class SubmissionSafetyTests(unittest.TestCase):
+    def test_official_routes_prefer_first_six_gpus(self) -> None:
+        target_gpus = [f"gpu-{index}" for index in range(1, 8)]
+        model = HFModelSummary(
+            repo_id="owner/model",
+            downloads=100,
+            last_modified=datetime(2026, 1, 1, tzinfo=timezone.utc),
+            pipeline_tag="text-generation",
+        )
+        candidates, skipped, failed = process_model_for_candidates(
+            model=model,
+            hf_discovery=SafeDiscovery(),  # type: ignore[arg-type]
+            modelhub_client=OtherGpuProcessedClient(),  # type: ignore[arg-type]
+            template_selector=TemplateSelector(),
+            target_gpus=target_gpus,
+            allowed_task_types=["text-generation"],
+            market_intelligence=AllGpuVettedMarket(),  # type: ignore[arg-type]
+            official_registry=AllGpuOfficialRegistry(),  # type: ignore[arg-type]
+        )
+        self.assertEqual(target_gpus[:6], [item["targetGpu"] for item in candidates])
+        self.assertEqual([], skipped)
+        self.assertEqual([], failed)
+
+    def test_official_routes_fall_back_after_preferred_six_are_unusable(self) -> None:
+        target_gpus = [f"gpu-{index}" for index in range(1, 8)]
+        model = HFModelSummary(
+            repo_id="owner/model",
+            downloads=100,
+            last_modified=datetime(2026, 1, 1, tzinfo=timezone.utc),
+            pipeline_tag="text-generation",
+        )
+        candidates, skipped, failed = process_model_for_candidates(
+            model=model,
+            hf_discovery=SafeDiscovery(),  # type: ignore[arg-type]
+            modelhub_client=OtherGpuProcessedClient(),  # type: ignore[arg-type]
+            template_selector=TemplateSelector(),
+            target_gpus=target_gpus,
+            allowed_task_types=["text-generation"],
+            market_intelligence=SeventhGpuFallbackMarket(),  # type: ignore[arg-type]
+            official_registry=AllGpuOfficialRegistry(),  # type: ignore[arg-type]
+        )
+        self.assertEqual(["gpu-7"], [item["targetGpu"] for item in candidates])
+        self.assertEqual(6, len(skipped))
+        self.assertEqual([], failed)
+
     def test_precheck_tracks_processed_gpus_without_blocking_other_gpus(self) -> None:
         payload = {
             "code": 0,