Azure Runtime Surfaces Audit¶

Scope¶

This audit captures the follow-up roadmap work that:

normalized Azure OpenAI and Azure AI Inference runtime configuration,
promoted the live Azure probes into first-class repo-rag and make commands, and
carried that work into the next step by adding a live repository-answer path that reuses local retrieval plus Azure-backed synthesis.

Executed successfully in this turn:

uv sync --extra azure
uv lock
make hooks-install
uv run python -m compileall src tests
uv run pytest tests/test_azure_runtime.py tests/test_workflow_live.py tests/test_utilities.py tests/test_cli_and_dspy.py tests/test_verification.py tests/test_repository_rag_bdd.py
uv run pytest tests/test_utilities.py tests/test_repository_rag_bdd.py
uv run repo-rag smoke-test
cargo build --manifest-path rust-cli/Cargo.toml
make quality
set -a; . /home/standard/dspy_rag_in_repo_docs_and_impl1/.env; set +a; make azure-openai-probe
set -a; . /home/standard/dspy_rag_in_repo_docs_and_impl1/.env; set +a; make azure-inference-probe
set -a; . /home/standard/dspy_rag_in_repo_docs_and_impl1/.env; set +a; make ask-live QUESTION="What does this repository research?" LIVE_PROVIDER=azure-openai
set -a; . /home/standard/dspy_rag_in_repo_docs_and_impl1/.env; set +a; make ask-live QUESTION="What does this repository research?" LIVE_PROVIDER=azure-inference

uv run pytest ...: passed, 32 focused tests covering Azure runtime normalization, the new live workflow path, CLI surfaces, utility surfaces, and Makefile verification
uv run pytest tests/test_utilities.py tests/test_repository_rag_bdd.py: passed, 10 tests
uv run repo-rag smoke-test: passed with answer_contains_repository: true, mcp_candidate_count: 1, and manifest path artifacts/azure/repo-rag-smoke.json
cargo build --manifest-path rust-cli/Cargo.toml: passed
make quality: passed with 99 tests and 85.19% total coverage
make azure-openai-probe: passed and returned OPENAI_OK from deployment gpt-4o, resolved model gpt-4o-2024-11-20, API version 2025-01-01-preview
make azure-inference-probe: passed and returned INFERENCE_OK, normalized the configured AZURE_INFERENCE_ENDPOINT from the full /chat/completions?... URI to the deployment-base endpoint /openai/deployments/gpt-4o, and resolved model gpt-4o-2024-11-20
make ask-live ... LIVE_PROVIDER=azure-openai: passed as a live round trip
make ask-live ... LIVE_PROVIDER=azure-inference: passed as a live round trip

Configured and verified in this turn:

Compile checks: present and passed through uv run python -m compileall src tests
Focused Azure/runtime/CLI utility tests: present and passed through the targeted uv run pytest slice above
Required utility pytest slice: present and passed through uv run pytest tests/test_utilities.py tests/test_repository_rag_bdd.py
Repository smoke test: present and passed through uv run repo-rag smoke-test
Rust build: present and passed through cargo build --manifest-path rust-cli/Cargo.toml
Lint, notebook lint, mypy, basedpyright, repository-surface verification, complexity, tests, and coverage: present and passed through make quality
Live Azure OpenAI probe surface: present and passed through make azure-openai-probe
Live Azure AI Inference probe surface: present and passed through make azure-inference-probe
Live repository-answer surface: present and executed successfully through make ask-live for both Azure providers

Still absent or not exercised in this turn:

UI or browser tests: none found in the repository configuration
Full notebook execution batch: notebook lint and repository-surface checks passed, but make notebook-report was not rerun end-to-end in this turn
Post-push GitHub Actions evidence: not yet available before the push for this change set

The new runtime loader accepts either a repo-local .env or already-exported process environment variables. In this clean worktree, .env was intentionally absent, so the live Azure commands were verified by exporting /home/standard/dspy_rag_in_repo_docs_and_impl1/.env into the process environment first.
The new make ask-live surface successfully reaches both Azure providers, but the quality of the returned answer is still constrained by the baseline lexical retriever. For the repository-scope question used here, both providers responded conservatively that the retrieved evidence did not explicitly answer the question, even though the baseline smoke test still confirms the local workflow can surface repository in its answer text.
The documentation now matches the code: the repo exposes ask-live, azure-openai-probe, and azure-inference-probe through both make and repo-rag, and the Azure AI Inference endpoint normalization is handled inside the package rather than only in ad hoc probe scripts.