v0.2.3:

Author: kazhou

.gitignore

@@ -19,6 +19,7 @@ __pycache__/
 # Distribution / packaging
 .Python
 build/
+!ui/frontend/src/app/build/
 develop-eggs/
 dist/
 downloads/

autochecklist/models.py

@@ -29,7 +29,7 @@ class ChecklistItem(BaseModel):
 
     id: str = Field(default_factory=lambda: str(uuid.uuid4())[:8])
     question: str
-    weight: float = Field(default=100.0, ge=0.0, le=100.0)  # RLCF uses 0-100
+    weight: float = 100.0  # RLCF uses 0-100; constraint omitted for cross-provider JSON-schema compat
     category: Optional[str] = None  # For grouping (e.g., "factuality", "format")
     metadata: Dict[str, Any] = Field(default_factory=dict)
 
@@ -263,7 +263,7 @@ class GeneratedQuestion(BaseModel):
 class GeneratedWeightedQuestion(BaseModel):
     """A single generated yes/no question with importance weight."""
     question: str
-    weight: int = Field(ge=0, le=100)
+    weight: int  # 0-100; constraint omitted for cross-provider JSON-schema compat (see ChecklistItem.weight)
 
 
 class ChecklistResponse(BaseModel):

autochecklist/refiners/base.py

@@ -75,16 +75,26 @@ def _call_model(
         prompt: str,
         system_prompt: Optional[str] = None,
         response_format: Optional[Dict] = None,
+        model: Optional[str] = None,
     ) -> str:
-        """Call the LLM and return the response text."""
+        """Call the LLM and return the response text.
+
+        Args:
+            prompt: User message content.
+            system_prompt: Optional system message.
+            response_format: Optional structured-output spec.
+            model: Optional model override. Defaults to ``self.model`` when
+                not provided. Useful when a refiner has a secondary model
+                role (e.g. ``Selector.classifier_model``).
+        """
         messages: List[Dict[str, str]] = []
         if system_prompt:
             messages.append({"role": "system", "content": system_prompt})
         messages.append({"role": "user", "content": prompt})
 
         client = self._get_or_create_client()
         kwargs: Dict[str, Any] = {
-            "model": self.model,
+            "model": model if model is not None else self.model,
             "messages": messages,
             "temperature": self.temperature,
             "max_tokens": 2048,

autochecklist/refiners/selector.py

@@ -281,7 +281,7 @@ def _classify_feedback(self, checklist: Checklist) -> Dict[int, Set[int]]:
                 questions=questions_text,
                 feedback_item=fb_item,
             )
-            response = self._call_model(prompt)
+            response = self._call_model(prompt, model=self.classifier_model)
 
             # Parse question numbers from response
             nums = re.findall(r"\d+", response)

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "autochecklist"
-version = "0.2.2"
+version = "0.2.3"
 description = "A library of checklist generation and scoring methods for LLM evaluation"
 authors = [{name = "ChicagoHAI"}]
 readme = "README.pypi.md"

tests/test_models/test_core_models.py

@@ -135,15 +135,14 @@ def test_weighted_checklist_response_valid(self):
         assert resp.questions[0].weight == 100
         assert resp.questions[1].weight == 50
 
-    def test_weighted_question_weight_bounds(self):
-        # Valid bounds
+    def test_weighted_question_accepts_full_int_range(self):
+        # Bounds intentionally not enforced: cross-provider JSON-schema compat
+        # (OpenAI/Anthropic/Gemini structured-output disagree on `minimum`/`maximum`).
+        # The 0-100 range is communicated via the generator prompt instead.
         GeneratedWeightedQuestion(question="test?", weight=0)
         GeneratedWeightedQuestion(question="test?", weight=100)
-        # Out of bounds
-        with pytest.raises(ValidationError):
-            GeneratedWeightedQuestion(question="test?", weight=-1)
-        with pytest.raises(ValidationError):
-            GeneratedWeightedQuestion(question="test?", weight=101)
+        GeneratedWeightedQuestion(question="test?", weight=-1)
+        GeneratedWeightedQuestion(question="test?", weight=101)
 
     def test_batch_scoring_response_valid(self):
         resp = BatchScoringResponse(answers=[

tests/test_refiners/test_selector.py

@@ -93,8 +93,32 @@ def test_refiner_name(self):
         selector = Selector()
         assert selector.refiner_name == "selector"
 
+    def test_default_classifier_model(self):
+        """Default classifier_model should be openai/gpt-4o-mini."""
+        selector = Selector()
+        assert selector.classifier_model == "openai/gpt-4o-mini"
+
+    def test_custom_classifier_model(self):
+        """Should accept custom classifier_model."""
+        selector = Selector(classifier_model="openai/gpt-5-nano")
+        assert selector.classifier_model == "openai/gpt-5-nano"
+
 
-class TestSelectorSmallChecklist:
+class TestSelectorClassifierModel:
+    """Tests that _classify_feedback honors classifier_model."""
+
+    def test_classify_uses_classifier_model_not_self_model(self, small_checklist):
+        """_classify_feedback should pass classifier_model, not self.model, to _call_model."""
+        selector = Selector(
+            model="anthropic/claude-sonnet-4.6",
+            classifier_model="openai/gpt-4o-mini",
+            observations=["obs A", "obs B"],
+        )
+        with patch.object(selector, "_call_model", return_value="0") as mock_call:
+            selector._classify_feedback(small_checklist)
+        assert mock_call.call_count == 2
+        for call in mock_call.call_args_list:
+            assert call.kwargs.get("model") == "openai/gpt-4o-mini"
     """Tests when checklist is already small enough."""
 
     @patch("autochecklist.refiners.selector.get_embeddings")

ui/frontend/src/app/build/page.tsx

@@ -0,0 +1,16 @@
+"use client";
+
+import { PageHeader } from "@/components/layout/PageHeader";
+import { DimensionForm } from "@/components/checklist_builder/DimensionForm";
+
+export default function BuildPage() {
+  return (
+    <div className="max-w-4xl mx-auto">
+      <PageHeader
+        title="Build a Checklist"
+        description="Define rubric dimensions to automatically build a checklist with DeductiveGenerator."
+      />
+      <DimensionForm />
+    </div>
+  );
+}