Release 0.7.11 routing safeguards

Author: anjieyang

openclaw-plugin/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@anjieyang/uncommon-route",
-  "version": "0.7.10",
+  "version": "0.7.11",
   "description": "OpenClaw plugin for UncommonRoute, the local LLM router that cuts premium-model spend",
   "type": "module",
   "main": "src/index.js",

openclaw-plugin/src/index.js

@@ -16,7 +16,7 @@
 import { spawn, execSync } from "node:child_process";
 import { setTimeout as sleep } from "node:timers/promises";
 
-const VERSION = "0.7.10";
+const VERSION = "0.7.11";
 const DEFAULT_PORT = 8403;
 const DEFAULT_UPSTREAM = "";
 const HEALTH_TIMEOUT_MS = 15_000;

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "uncommon-route"
-version = "0.7.10"
+version = "0.7.11"
 description = "Local LLM router that cuts premium-model spend with adaptive 3-tier routing, OpenAI + Anthropic compatible"
 requires-python = ">=3.11"
 license = "MIT"

tests/test_model_experience.py

@@ -71,6 +71,17 @@ def test_infer_capabilities_marks_gpt5_family_as_reasoning() -> None:
     assert capabilities.reasoning is True
 
 
+def test_infer_capabilities_does_not_mark_image_generation_as_tool_chat() -> None:
+    capabilities = infer_capabilities(
+        "google/gemini-3-pro-image-preview",
+        ModelPricing(2.0, 12.0),
+        has_explicit_pricing=True,
+    )
+
+    assert capabilities.vision is True
+    assert capabilities.tool_calling is False
+
+
 def test_invalid_upstream_pricing_is_not_treated_as_free() -> None:
     pricing = _parse_upstream_pricing({"prompt": "not-a-number", "completion": "0"})
     capabilities = infer_capabilities(
@@ -429,6 +440,51 @@ def test_best_mode_uses_higher_quality_threshold() -> None:
         "Same model should have same predicted quality regardless of mode"
 
 
+def test_auto_mode_cost_sanity_guard_blocks_dominated_expensive_exploration(monkeypatch) -> None:
+    """AUTO exploration should not pick pricier same-quality peers with weaker priors."""
+    import uncommon_route.benchmark as benchmark
+    import uncommon_route.router.selector as selector
+
+    class DummyBenchmarkCache:
+        def get_all_qualities(self, models):
+            return {
+                "anthropic/claude-opus-4-7": 0.84,
+                "google/gemini-3.1-pro-preview": 0.88,
+            }
+
+    monkeypatch.setattr(benchmark, "get_benchmark_cache", lambda: DummyBenchmarkCache())
+    draws = iter([0.99, 0.10])
+    monkeypatch.setattr(selector._rng, "betavariate", lambda _alpha, _beta: next(draws))
+
+    pricing = {
+        "anthropic/claude-opus-4-7": ModelPricing(5.0, 25.0),
+        "google/gemini-3.1-pro-preview": ModelPricing(2.0, 12.0),
+    }
+    capabilities = {
+        model: infer_capabilities(model, model_pricing, has_explicit_pricing=True)
+        for model, model_pricing in pricing.items()
+    }
+
+    decision = select_from_pool(
+        complexity=0.90,
+        mode=RoutingMode.AUTO,
+        confidence=0.9,
+        reasoning_text="test",
+        available_models=list(pricing),
+        estimated_input_tokens=4_000,
+        max_output_tokens=800,
+        prompt="Solve a hard reasoning bug.",
+        pricing=pricing,
+        capabilities=capabilities,
+        requirements=RequestRequirements(prefers_reasoning=True),
+        selection_weights=get_selection_weights(DEFAULT_CONFIG, RoutingMode.AUTO),
+        bandit_config=BanditConfig(enabled=True, enabled_tiers=(Tier.COMPLEX,)),
+    )
+
+    assert decision.model == "google/gemini-3.1-pro-preview"
+    assert "cost-sanity=anthropic/claude-opus-4-7->google/gemini-3.1-pro-preview" in decision.reasoning
+
+
 def test_feedback_collector_updates_model_experience() -> None:
     store = ModelExperienceStore(storage=InMemoryModelExperienceStorage())
     feedback = FeedbackCollector(model_experience=store)

tests/test_quality_routing.py

@@ -57,6 +57,16 @@ def test_model_served_quality_normalizes_moonshot_provider_aliases() -> None:
     assert model_served_quality("moonshotai/kimi-k2.5", CapabilityLane.GENERAL, caps) is ServedQuality.BALANCED
 
 
+def test_openai_nano_is_not_premium_for_reasoning_lane() -> None:
+    caps = ModelCapabilities(tool_calling=True, reasoning=True)
+
+    assert model_served_quality(
+        "openai/gpt-5.4-nano-2026-03-17",
+        CapabilityLane.REASONING,
+        caps,
+    ) is ServedQuality.ECONOMY
+
+
 def test_quality_guards_keep_auto_complex_at_balanced_floor_without_opus_only_pool() -> None:
     caps = _caps()
     guard = apply_quality_guards(

tests/test_reported_issues.py

@@ -420,6 +420,9 @@ def test_issue6_legacy_model_ids_resolve_to_live_successors():
 
     live_models = [
         "x-ai/grok-4-1-fast-non-reasoning",
+        "x-ai/grok-4.1-fast-reasoning",
+        "x-ai/grok-code-fast-1",
+        "google/gemini-3.1-pro-preview",
         "openai/gpt-4.1",
         "openai/gpt-5.3-codex",
         "openai/gpt-5.4-mini-2026-03-17",
@@ -441,6 +444,11 @@ def test_issue6_legacy_model_ids_resolve_to_live_successors():
     mapper._discovered = True
 
     assert mapper.resolve("xai/grok-4-0709") == "x-ai/grok-4-1-fast-non-reasoning"
+    assert mapper.resolve("xai/grok-4-1-fast-reasoning") == "x-ai/grok-4.1-fast-reasoning"
+    assert mapper.resolve("xai/grok-4-1-fast-non-reasoning") == "x-ai/grok-4-1-fast-non-reasoning"
+    assert mapper.resolve("xai/grok-code-fast-1") == "x-ai/grok-code-fast-1"
+    assert mapper.resolve("google/gemini-3-pro-preview") == "google/gemini-3.1-pro-preview"
+    assert mapper.resolve("google/gemini-3.1-pro") == "google/gemini-3.1-pro-preview"
     assert mapper.resolve("openai/gpt-4o") == "openai/gpt-4.1"
     assert mapper.resolve("openai/gpt-5.2-codex") == "openai/gpt-5.3-codex"
     assert mapper.resolve("openai/o1-mini") == "openai/gpt-5.4-mini-2026-03-17"
@@ -450,6 +458,11 @@ def test_issue6_legacy_model_ids_resolve_to_live_successors():
         model in mapper.unresolved_models()
         for model in (
             "xai/grok-4-0709",
+            "xai/grok-4-1-fast-reasoning",
+            "xai/grok-4-1-fast-non-reasoning",
+            "xai/grok-code-fast-1",
+            "google/gemini-3-pro-preview",
+            "google/gemini-3.1-pro",
             "openai/gpt-4o",
             "openai/gpt-5.2-codex",
             "openai/o1-mini",
@@ -459,6 +472,38 @@ def test_issue6_legacy_model_ids_resolve_to_live_successors():
     )
 
 
+def test_issue6_image_generation_models_do_not_enter_chat_routing_pool():
+    """Image-generation catalog entries should not become chat candidates."""
+    from uncommon_route.model_map import DiscoveredModel, ModelMapper, is_routable_chat_model
+    from uncommon_route.router.types import ModelCapabilities, ModelPricing
+
+    mapper = ModelMapper("https://api.commonstack.ai/v1")
+    for model_id in [
+        "google/gemini-3-pro-image-preview",
+        "google/gemini-3.1-flash-image-preview",
+        "google/gemini-3.1-pro-preview",
+        "minimax/minimax-m2.7",
+    ]:
+        provider = model_id.split("/", 1)[0]
+        mapper._pool[model_id] = DiscoveredModel(
+            id=model_id,
+            provider=provider,
+            owned_by=provider,
+            pricing=ModelPricing(0.2, 0.8),
+            capabilities=ModelCapabilities(tool_calling=True, vision=True),
+        )
+        mapper._upstream_models.add(model_id)
+    mapper._build_map()
+    mapper._discovered = True
+
+    assert not is_routable_chat_model("google/gemini-3-pro-image-preview")
+    assert mapper.resolve("google/gemini-3-pro-preview") == "google/gemini-3.1-pro-preview"
+    assert "google/gemini-3-pro-image-preview" in mapper.available_models
+    assert "google/gemini-3-pro-image-preview" not in mapper.routable_models
+    assert "google/gemini-3.1-flash-image-preview" not in mapper.routable_models
+    assert "google/gemini-3.1-pro-preview" in mapper.routable_models
+
+
 # ─── Summary: combined issue reproduction ───
 
 def test_combined_claude_code_session_simulation():

tests/test_v2_routing_logic.py

@@ -0,0 +1,34 @@
+from __future__ import annotations
+
+from uncommon_route.router import api
+from uncommon_route.router.types import RoutingFeatures
+from uncommon_route.signals.base import TierVote
+
+
+class _FakeSignal:
+    def __init__(self, vote: TierVote) -> None:
+        self.vote = vote
+        self._embed_fn = None
+
+    def predict(self, row: dict) -> TierVote:
+        return self.vote
+
+
+def test_soft_structural_floor_keeps_short_implementation_out_of_economy(monkeypatch) -> None:
+    """A short implementation ask should not be flattened to SIMPLE by A+C."""
+    monkeypatch.setattr(api, "_ensure_v2_signals", lambda: None)
+    monkeypatch.setattr(api, "_v2_sig_a", _FakeSignal(TierVote(0, 0.75)))
+    monkeypatch.setattr(api, "_v2_sig_b", _FakeSignal(TierVote(3, 0.70)))
+    monkeypatch.setattr(api, "_v2_sig_c", _FakeSignal(TierVote(0, 0.44)))
+    monkeypatch.setattr(api, "_v2_calibrator", None)
+
+    result = api._v2_classify(
+        "Implement a Python CLI app with tests, packaging, error handling, and documentation.",
+        system_prompt=None,
+        messages=None,
+        routing_features=RoutingFeatures(),
+        context_features=None,
+    )
+
+    assert result.tier_id == 1
+    assert "v2:structural-medium-floor" in result.signals_text

uncommon_route/cli.py

@@ -773,7 +773,10 @@ async def _check_provider(base_url: str, api_key: str) -> bool:
         async with httpx.AsyncClient(timeout=httpx.Timeout(5.0, connect=3.0)) as client:
             resp = await client.get(
                 f"{base_url.rstrip('/')}/models",
-                headers={"authorization": f"Bearer {api_key}"},
+                headers={
+                    "authorization": f"Bearer {api_key}",
+                    "user-agent": "uncommon-route/provider-check",
+                },
             )
             return resp.status_code == 200
     except Exception:  # noqa: BLE001

uncommon_route/model_map.py

@@ -83,9 +83,25 @@ def detect_provider(url: str) -> tuple[str, bool]:
     "google/gemini-2.5-flash-lite": [
         "google/gemini-3.1-flash-lite-preview",
     ],
+    "google/gemini-3-pro-preview": [
+        "google/gemini-3.1-pro-preview",
+    ],
+    "google/gemini-3.1-pro": [
+        "google/gemini-3.1-pro-preview",
+    ],
     "xai/grok-4-0709": [
         "x-ai/grok-4-1-fast-non-reasoning",
     ],
+    "xai/grok-4-1-fast-reasoning": [
+        "x-ai/grok-4.1-fast-reasoning",
+        "x-ai/grok-4-1-fast-reasoning",
+    ],
+    "xai/grok-4-1-fast-non-reasoning": [
+        "x-ai/grok-4-1-fast-non-reasoning",
+    ],
+    "xai/grok-code-fast-1": [
+        "x-ai/grok-code-fast-1",
+    ],
     "openai/gpt-4o": [
         "openai/gpt-4.1",
     ],
@@ -139,6 +155,28 @@ def _parse_upstream_pricing(raw: dict | None) -> ModelPricing:
 
 _REASONING_POSITIVE = ("reason", "r1", "thinking", "think", "o1-", "o3", "o4-", "gpt-5")
 _REASONING_NEGATIVE = ("non-reason", "non_reason", "no-reason", "non-thinking")
+_NON_CHAT_MODEL_MARKERS = (
+    "-image",
+    "image-",
+    "-video",
+    "video-",
+    "-audio",
+    "audio-",
+    "speech",
+    "tts",
+    "embedding",
+    "embed",
+)
+
+
+def is_routable_chat_model(model_id: str) -> bool:
+    """Return False for modality-specific generation models in /v1/models.
+
+    Gateways often list image, video, audio, and embedding models beside chat
+    models. The chat router should not select those for text/tool requests.
+    """
+    core = _core(str(model_id or "").lower())
+    return not any(marker in core for marker in _NON_CHAT_MODEL_MARKERS)
 
 
 def infer_capabilities(
@@ -163,6 +201,8 @@ def infer_capabilities(
     if provider == "anthropic" and "opus" in core:
         reasoning = True
 
+    image_generation = not is_routable_chat_model(model_id) and "image" in core
+
     vision = False
     if any(p in core for p in ("-vl", "vision", "image")):
         vision = True
@@ -171,7 +211,7 @@ def infer_capabilities(
     if provider == "openai" and any(p in core for p in ("4o", "gpt-5", "gpt-4")):
         vision = True
 
-    tool_calling = True
+    tool_calling = not image_generation
 
     free = (
         has_explicit_pricing
@@ -367,6 +407,8 @@ def _fuzzy_match(self, internal: str) -> str | None:
         best: str | None = None
 
         for upstream in self._upstream_models:
+            if is_routable_chat_model(internal) and not is_routable_chat_model(upstream):
+                continue
             up_core = _core(upstream)
             up_norm = _normalize(up_core)
             up_prov = _provider_prefix(upstream)
@@ -459,6 +501,11 @@ def available_models(self) -> list[str]:
         """All available upstream model IDs, sorted."""
         return sorted(self._pool.keys())
 
+    @property
+    def routable_models(self) -> list[str]:
+        """Chat-routable upstream model IDs, sorted."""
+        return [model for model in self.available_models if is_routable_chat_model(model)]
+
     def get_pricing(self, model_id: str) -> ModelPricing | None:
         """Look up pricing for a single model (internal or upstream ID)."""
         if model_id in self._pool:

uncommon_route/providers.py

@@ -196,7 +196,10 @@ def verify_key(base_url: str, api_key: str) -> tuple[bool, str]:
     try:
         resp = httpx.get(
             url,
-            headers={"authorization": f"Bearer {api_key}"},
+            headers={
+                "authorization": f"Bearer {api_key}",
+                "user-agent": "uncommon-route/provider-check",
+            },
             timeout=httpx.Timeout(8.0, connect=4.0),
         )
         if resp.status_code == 200: