feat: add Inspect AI integration package (learning-commons-inspect-scorers)

.github/workflows/publish-inspect-python.yml

@@ -0,0 +1,59 @@
+# ─── Before the first publish (integrations-inspect-python-v0.1.0) ───────────
+#
+# 1. Tighten the SDK dependency. integrations/inspect-python/pyproject.toml pins
+#    `learning-commons-evaluators>=0.2.0`, but the integration requires the
+#    LLMGeneratorProtocol / llm_provider support that ships in SDK 0.3.0. Raise
+#    the floor to `>=0.3.0` once 0.3.0 is on PyPI — and do not publish this
+#    package before then, or installs resolve a SDK without llm_provider.
+#
+# 2. Remove `release-as: "0.1.0"` from the integrations/inspect-python block in
+#    release-please-config.json IMMEDIATELY after 0.1.0 lands on PyPI. It is a
+#    sticky pin: every later release-please PR stays at 0.1.0 until removed, and
+#    `skip-existing: true` below silently hides the duplicate-version rejection.
+#
+# ─────────────────────────────────────────────────────────────────────────────
+
+name: Publish Inspect integration
+
+on:
+  release:
+    types: [published]
+  workflow_dispatch:
+
+jobs:
+  publish:
+    if: github.event_name == 'workflow_dispatch' || startsWith(github.event.release.tag_name, 'integrations-inspect-python-v')
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+    defaults:
+      run:
+        working-directory: integrations/inspect-python
+    concurrency:
+      group: publish-inspect-python
+      cancel-in-progress: false
+    steps:
+      - name: ⬇️ Checkout repo
+        uses: actions/checkout@v6
+
+      - name: 🐍 Setup Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.13"
+
+      - name: 📥 Install build tooling
+        run: python -m pip install --upgrade build twine
+
+      - name: 📦 Build sdist + wheel
+        run: python -m build
+
+      - name: 🧪 Validate distribution metadata
+        run: python -m twine check dist/*
+
+      - name: 🚀 Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1
+        with:
+          packages-dir: integrations/inspect-python/dist
+          skip-existing: true

.github/workflows/test-inspect-python.yml

@@ -0,0 +1,63 @@
+name: 🔍 Test Inspect integration
+
+# Runs only when the integration package itself changes, or when the Python SDK
+# it depends on changes (a SDK change — including a version bump — can break the
+# integration, so we re-verify here too).
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - "integrations/inspect-python/**"
+      - "sdks/python/**"
+      - "sdks/settings/**"
+      - ".github/workflows/test-inspect-python.yml"
+  pull_request:
+    paths:
+      - "integrations/inspect-python/**"
+      - "sdks/python/**"
+      - "sdks/settings/**"
+      - ".github/workflows/test-inspect-python.yml"
+
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  verify:
+    name: Verify (Python ${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.10", "3.11", "3.12", "3.13"]
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v6
+
+      - name: Set up Python ${{ matrix.python-version }}
+        uses: actions/setup-python@v6
+        with:
+          python-version: ${{ matrix.python-version }}
+
+      - name: Install in-repo SDK + integration package
+        # Install the SDK from source first: the integration relies on
+        # LLMGeneratorProtocol (SDK >=0.3.0), which is not yet on PyPI. The
+        # editable install satisfies the >=0.2.0 floor while providing the
+        # unreleased protocol code, so CI tests against current monorepo state.
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e ./sdks/python
+          pip install -e "./integrations/inspect-python[dev]"
+
+      - name: Lint
+        working-directory: integrations/inspect-python
+        run: |
+          python -m ruff check .
+          python -m ruff format --check .
+
+      - name: Test
+        working-directory: integrations/inspect-python
+        run: python -m pytest

.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 +1,6 @@
 {
   "evals/prompts": "1.5.0",
   "sdks/python": "0.2.0",
-  "sdks/typescript": "0.7.0"
+  "sdks/typescript": "0.7.0",
+  "integrations/inspect-python": "0.1.0"
 }

integrations/inspect-python/.gitignore

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

integrations/inspect-python/CHANGELOG.md

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

integrations/inspect-python/README.md

@@ -0,0 +1,82 @@
+# learning-commons-inspect-scorers
+
+[Inspect AI](https://inspect.aisi.org.uk/) scorer wrappers for the [Learning Commons evaluators](https://github.com/learning-commons-org/evaluators) SDK.
+
+## Installation
+
+```bash
+pip install learning-commons-inspect-scorers
+```
+
+> **Note:** Requires `learning-commons-evaluators>=0.2.0`. During local development
+> install the SDK from the repo root first:
+> ```bash
+> pip install -e sdks/python
+> pip install -e integrations/inspect-python
+> ```
+
+## Usage
+
+### Grade Level Appropriateness scorer
+
+Evaluates whether model output (or generated artifact files) is written at the
+appropriate reading level for a target K-12 grade band.
+
+```python
+from inspect_ai import Task, task
+from inspect_ai.dataset import csv_dataset, FieldSpec
+from inspect_ai.solver import generate
+from learning_commons_inspect_scorers import gla_scorer
+
+@task
+def my_eval():
+    return Task(
+        dataset=csv_dataset("samples.csv"),  # requires target_grade column
+        solver=[generate()],
+        scorer=gla_scorer(),
+    )
+```
+
+The dataset CSV must include a `target_grade` metadata column with one of:
+`K-1`, `2-3`, `4-5`, `6-8`, `9-10`, `11-CCR`.
+
+### Scoring text your task produced (not the completion)
+
+By default the scorer grades `state.output.completion`. To grade text from
+somewhere else — files your solver wrote, a specific message, filtered content —
+pass a `text_fn`. This keeps task-specific knowledge (file layout, naming,
+formatting) in your task, not in this package:
+
+```python
+def student_artifact_text(state):
+    # read whatever your task produced; return None to skip the sample
+    return "\n\n".join(read_my_files(state)) or None
+
+scorer=gla_scorer(text_fn=student_artifact_text)
+```
+
+### Re-scoring an existing log from the CLI
+
+Once installed, scorers are registered via Inspect's entry point system:
+
+```bash
+inspect score logs/my-eval.eval --scorer learning_commons_inspect_scorers/gla_scorer
+```
+
+## Configuration
+
+| Parameter | Default | Description |
+|---|---|---|
+| `grader_model` | `"anthropic/claude-opus-4-8"` | Inspect model string for the grading LLM. Uses Inspect's model system — no separate API key configuration needed. |
+| `target_grade_key` | `"target_grade"` | Metadata key holding the expected grade band. |
+| `allow_adjacent` | `True` | If `True`, the one grade band above or below the target also passes. |
+| `text_fn` | `None` | Callable `(TaskState) -> str \| None` returning the text to grade. Defaults to `state.output.completion`. Return `None`/empty to skip the sample. Caller keeps text within the evaluator's input-length limit. |
+
+## Development
+
+```bash
+# From repo root
+pip install -e sdks/python
+pip install -e "integrations/inspect-python[dev]"
+pytest integrations/inspect-python/tests/
+```

integrations/inspect-python/pyproject.toml

@@ -0,0 +1,75 @@
+[build-system]
+requires = ["setuptools>=61", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "learning-commons-inspect-scorers"
+version = "0.1.0"
+description = "Inspect AI scorer wrappers for Learning Commons evaluators"
+readme = "README.md"
+license = { text = "MIT" }
+requires-python = ">=3.10"
+authors = [{ name = "Learning Commons" }]
+keywords = ["education", "evaluators", "inspect", "evals", "scoring"]
+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",
+    # Score.unscored() — used in every skip/error path of gla_scorer — was added in inspect-ai 0.3.214.
+    "inspect-ai>=0.3.214",
+]
+
+[project.optional-dependencies]
+dev = [
+    "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/inspect-python"
+Documentation = "https://docs.learningcommons.org/evaluators"
+"Bug Tracker" = "https://github.com/learning-commons-org/evaluators/issues"
+
+# Registers scorers with Inspect's component discovery via setuptools entry points.
+# Once installed, scorers are accessible as e.g. `learning_commons_inspect_scorers/gla_scorer`
+# from the CLI: inspect score log.eval --scorer learning_commons_inspect_scorers/gla_scorer
+[project.entry-points.inspect_ai]
+learning_commons_inspect_scorers = "learning_commons_inspect_scorers._registry"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[tool.setuptools.package-data]
+learning_commons_inspect_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
+plugins = ["pydantic.mypy"]
+warn_unused_configs = true
+show_error_codes = true

integrations/inspect-python/src/learning_commons_inspect_scorers/__init__.py

@@ -0,0 +1,6 @@
+"""Learning Commons Inspect scorers — Inspect AI wrappers for LC evaluators."""
+
+from learning_commons_inspect_scorers.adapter import InspectModelAdapter
+from learning_commons_inspect_scorers.gla import gla_scorer
+
+__all__ = ["InspectModelAdapter", "gla_scorer"]

integrations/inspect-python/src/learning_commons_inspect_scorers/_registry.py

@@ -0,0 +1,10 @@
+"""Entry point registry — imported by Inspect via the inspect_ai setuptools entry point.
+
+Importing this module registers all scorers with Inspect's component system,
+making them accessible by name (e.g. learning_commons_inspect_scorers/gla_scorer)
+from both the Python API and the CLI.
+"""
+
+from learning_commons_inspect_scorers.gla import (
+    gla_scorer,  # noqa: F401 — import triggers @scorer registry side-effect
+)