TODO

This backlog is generated from todo-backlog.yaml via make todo-sync.

🎯	🧭 Area	📌 TODO	🔗 Primary Surfaces	✅ Done When
🔎	Retrieval, Corpus, And Evaluation	Strengthen retrieval quality beyond the lexical baseline.	src/repo_rag_lab/retrieval.py src/repo_rag_lab/workflow.py src/repo_rag_lab/dspy_workflow.py	A stronger retrieval layer replaces simple lexical overlap. Direct file retrieval and MCP-powered retrieval are compared explicitly. Retrieval upgrades are shared by both the baseline and DSPy paths.
🧪	Retrieval, Corpus, And Evaluation	Expand benchmark and training evidence.	samples/training/repository_training_examples.yaml src/repo_rag_lab/benchmarks.py tests/test_benchmarks_and_notebook_scaffolding.py	More training examples carry expected_sources evidence. Repository-specific benchmark questions and expected answers are expanded. Regression and artifact-persistence tests land once training exists.
🗺	Retrieval, Corpus, And Evaluation	Keep the benchmark corpus definition stable as the repository grows.	src/repo_rag_lab/benchmarks.py docs/audit/README.md samples/logs/	Operational evidence surfaces stay excluded from the benchmark corpus. Benchmark exclusions are reviewed whenever new generated docs or artifacts appear. Assertions keep measuring repository content instead of verification byproducts.
📚	Retrieval, Corpus, And Evaluation	Continue improving corpus planning.	src/repo_rag_lab/population_samples.py samples/population/repository_population_candidates.yaml notebooks/04_sample_population_lab.ipynb	Benchmark hit data drives reranking more systematically. Population-candidate discovery expands when new repository surfaces matter.
🧾	Notebook And Documentation Surfaces	Keep notebooks executable as research playbooks.	notebooks/01_repo_rag_research.ipynb notebooks/03_dspy_training_lab.ipynb src/repo_rag_lab/notebook_scaffolding.py	Notebook execution checks land in CI if notebook execution becomes an acceptance surface. Notebook cell IDs and repository-surface validation stay intact as scaffolding evolves. Notebook logging and tuning metadata outputs stay documented under artifacts/.
📝	Notebook And Documentation Surfaces	Keep the DSPy and utility documentation current.	README.DSPY.MD utilities/README.md README.md	README.DSPY.MD is refreshed whenever training or benchmark workflows change. README.md, utilities/README.md, and notebook guidance match the actual CLI and Make targets.
🚦	Verification, CI, And Deployment	Expand verification beyond the current smoke and BDD emphasis.	tests/test_repository_rag_bdd.py documentation/azure-deployment.md tests/fixtures/hushwheel_lexiconarium/	A live or mocked Azure inference integration test exists. Integration-oriented tests are split out or labeled clearly as the repository grows. Future DSPy training and artifact-loading paths have dedicated tests.
🧭	Verification, CI, And Deployment	Keep CI current with platform changes.	.github/workflows/ci.yml samples/logs/ docs/audit/README.md	GitHub Actions dependencies stop emitting Node.js 20 deprecation warnings. Post-push CI evidence stays current under samples/logs/.
📦	Verification, CI, And Deployment	Keep packaging and publish surfaces healthy.	.github/workflows/publish.yml pyproject.toml publication/README.md	The Publish workflow is re-checked once versioned DSPy artifacts or release tags exist.
🦀	Repo Hygiene And Workflow Policy	Keep the committed rust-cli/Cargo.lock policy intentional.	rust-cli/Cargo.lock REPO_COMPLETENESS_CHECKLIST.md rust-cli/Cargo.toml	The lockfile is refreshed only when Rust dependencies change. Lockfile drift is treated as repository-surface work rather than incidental noise.
🧹	Repo Hygiene And Workflow Policy	Keep audit notes and backlog documents synchronized.	docs/audit/README.md TODO.MD samples/logs/	Audit notes are updated whenever verification status changes materially. The backlog is refreshed when roadmap items are completed, replaced, or reprioritized.

Edit todo-backlog.yaml, then run make todo-sync to refresh this table and the publication-ready LaTeX include.