feat: add Arize, Langfuse, and Braintrust integration packages

.gitignore

@@ -2,6 +2,15 @@
 .venv/
 __pycache__/
 *.pyc
+*.egg-info/
+dist/
+build/
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+
+# Inspect AI eval logs
+logs/
 
 # Jupyter Notebook
 .ipynb_checkpoints/

.release-please-manifest.json

@@ -1,5 +1,8 @@
 {
   "evals/prompts": "1.5.0",
   "sdks/python": "0.2.0",
-  "sdks/typescript": "0.7.0"
+  "sdks/typescript": "0.7.0",
+  "integrations/langfuse-python": "0.1.0",
+  "integrations/arize-python": "0.1.0",
+  "integrations/braintrust-python": "0.1.0"
 }

integrations/arize-python/.gitignore

@@ -0,0 +1,6 @@
+*.egg-info/
+dist/
+build/
+__pycache__/
+.pytest_cache/
+.mypy_cache/

integrations/arize-python/CHANGELOG.md

@@ -0,0 +1 @@
+# Changelog

integrations/arize-python/README.md

@@ -0,0 +1,45 @@
+# learning-commons-arize-scorers
+
+[Arize/Phoenix](https://phoenix.arize.com/) OTel tracing adapter for the [Learning Commons evaluators](https://github.com/learning-commons-org/evaluators) SDK.
+
+Wraps any `LLMGeneratorProtocol` adapter and emits [OpenInference](https://github.com/Arize-ai/openinference) spans compatible with Arize Phoenix and any OTel backend.
+
+## Installation
+
+```bash
+pip install learning-commons-arize-scorers
+```
+
+## Usage
+
+```python
+from learning_commons_arize_scorers import PhoenixTracingAdapter
+from learning_commons_inspect_scorers.adapter import InspectModelAdapter
+from learning_commons_evaluators import GradeLevelAppropriatenessEvaluator
+from learning_commons_evaluators.config import create_config_no_telemetry
+
+adapter = PhoenixTracingAdapter(
+    InspectModelAdapter("anthropic/claude-opus-4-8"),
+    capture_message_content=False,  # False by default — K-12 privacy
+)
+evaluator = GradeLevelAppropriatenessEvaluator(
+    config=create_config_no_telemetry(),
+    llm_provider=adapter,
+)
+```
+
+## Configuration
+
+| Parameter | Default | Description |
+|---|---|---|
+| `inner` | required | Any `LLMGeneratorProtocol` adapter to wrap. |
+| `tracer` | auto | OTel `Tracer`. Defaults to `trace.get_tracer("learning_commons_arize_scorers")`. |
+| `capture_message_content` | `False` | Set `True` to include prompt/response text in spans. Off by default — student data may be sensitive. |
+
+## Development
+
+```bash
+pip install -e sdks/python
+pip install -e "integrations/arize-python[dev]"
+pytest integrations/arize-python/tests/
+```

integrations/arize-python/pyproject.toml

@@ -0,0 +1,68 @@
+[build-system]
+requires = ["setuptools>=61", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "learning-commons-arize-scorers"
+version = "0.1.0"
+description = "Arize/Phoenix OTel tracing adapter for Learning Commons evaluators"
+readme = "README.md"
+license = { text = "MIT" }
+requires-python = ">=3.10"
+authors = [{ name = "Learning Commons" }]
+keywords = ["education", "evaluators", "arize", "phoenix", "opentelemetry", "tracing"]
+classifiers = [
+    "Development Status :: 3 - Alpha",
+    "Intended Audience :: Developers",
+    "License :: OSI Approved :: MIT License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Education",
+]
+dependencies = [
+    "learning-commons-evaluators>=0.2.0",
+    "opentelemetry-api>=1.0.0",
+]
+
+[project.optional-dependencies]
+dev = [
+    "opentelemetry-sdk>=1.0.0",
+    "pytest>=7.0.0",
+    "pytest-asyncio>=0.21.0",
+    "ruff>=0.9.0",
+    "mypy>=1.14.0",
+]
+
+[project.urls]
+Homepage = "https://github.com/learning-commons-org/evaluators"
+Repository = "https://github.com/learning-commons-org/evaluators/tree/main/integrations/arize-python"
+Documentation = "https://docs.learningcommons.org/evaluators"
+"Bug Tracker" = "https://github.com/learning-commons-org/evaluators/issues"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+learning_commons_arize_scorers = ["py.typed"]
+
+[tool.pytest.ini_options]
+asyncio_mode = "auto"
+testpaths = ["tests"]
+
+[tool.ruff]
+target-version = "py310"
+line-length = 100
+
+[tool.ruff.lint]
+select = ["E", "W", "F", "I", "UP", "B", "SIM"]
+ignore = ["E501"]
+
+[tool.mypy]
+python_version = "3.10"
+mypy_path = ["src", "tests"]
+explicit_package_bases = true
+warn_unused_configs = true
+show_error_codes = true

integrations/arize-python/src/learning_commons_arize_scorers/__init__.py

@@ -0,0 +1,5 @@
+"""Learning Commons Arize scorers — OpenInference OTel tracing adapter for LC evaluators."""
+
+from learning_commons_arize_scorers.adapter import PhoenixTracingAdapter
+
+__all__ = ["PhoenixTracingAdapter"]

integrations/arize-python/src/learning_commons_arize_scorers/adapter.py

@@ -0,0 +1,79 @@
+"""PhoenixTracingAdapter — decorates any LLMGeneratorProtocol with OpenInference OTel spans."""
+
+from __future__ import annotations
+
+from opentelemetry import trace
+from opentelemetry.trace import Tracer
+from opentelemetry.trace.status import Status, StatusCode
+
+from learning_commons_evaluators.schemas.llm_provider import (
+    GenerateConfig,
+    LLMGeneratorProtocol,
+    LLMResponse,
+)
+
+
+class PhoenixTracingAdapter:
+    """Decorator adapter: wraps any LLMGeneratorProtocol, emits OpenInference OTel spans.
+
+    Composes with any other adapter::
+
+        from learning_commons_arize_scorers import PhoenixTracingAdapter
+        from learning_commons_inspect_scorers.adapter import InspectModelAdapter
+
+        adapter = PhoenixTracingAdapter(InspectModelAdapter("anthropic/claude-opus-4-8"))
+        evaluator = GradeLevelAppropriatenessEvaluator(config=..., llm_provider=adapter)
+
+    Args:
+        inner: The underlying adapter to delegate generation to.
+        tracer: OTel Tracer instance. Defaults to a tracer named
+                ``"learning_commons_arize_scorers"``.
+        capture_message_content: If ``True``, writes system and human prompt text
+                and the model response into span attributes. Defaults to ``False``.
+
+                .. warning::
+                    Enabling this may capture student-submitted text and other PII
+                    into your observability backend. Ensure your data handling
+                    controls (FERPA, COPPA for K-12) permit this before enabling.
+    """
+
+    def __init__(
+        self,
+        inner: LLMGeneratorProtocol,
+        tracer: Tracer | None = None,
+        *,
+        capture_message_content: bool = False,
+    ) -> None:
+        self._inner = inner
+        self._tracer = tracer or trace.get_tracer("learning_commons_arize_scorers")
+        self._capture_message_content = capture_message_content
+
+    async def generate(
+        self, *, system: str, human: str, config: GenerateConfig | None = None
+    ) -> LLMResponse:
+        with self._tracer.start_as_current_span("llm.generate") as span:
+            span.set_attribute("openinference.span.kind", "LLM")
+            span.set_attribute("gen_ai.operation.name", "chat")
+            if self._capture_message_content:
+                span.set_attribute("llm.input_messages.0.message.role", "system")
+                span.set_attribute("llm.input_messages.0.message.content", system)
+                span.set_attribute("llm.input_messages.1.message.role", "user")
+                span.set_attribute("llm.input_messages.1.message.content", human)
+            try:
+                response = await self._inner.generate(system=system, human=human, config=config)
+                span.set_attribute("gen_ai.response.model", response.model)
+                span.set_attribute("llm.model_name", response.model)
+                if response.input_tokens is not None:
+                    span.set_attribute("gen_ai.usage.input_tokens", response.input_tokens)
+                    span.set_attribute("llm.token_count.prompt", response.input_tokens)
+                if response.output_tokens is not None:
+                    span.set_attribute("gen_ai.usage.output_tokens", response.output_tokens)
+                    span.set_attribute("llm.token_count.completion", response.output_tokens)
+                if self._capture_message_content:
+                    span.set_attribute("llm.output_messages.0.message.role", "assistant")
+                    span.set_attribute("llm.output_messages.0.message.content", response.content)
+                return response
+            except Exception as exc:
+                span.record_exception(exc)
+                span.set_status(Status(StatusCode.ERROR))
+                raise