Roadmap¶
Planned improvements to this list and its automation. This is a living document: items are unordered within each horizon, carry no date commitments, and may change as the ecosystem evolves. Contributions toward any item are welcome — see CONTRIBUTING.md. Completed work is recorded in the Changelog.
Near-term¶
- Re-verify the benchmark rows the weekly audit still flags as stale — the vendor figures in §1 Vector Databases, §4 Caching, and §7 Reliability / SLA predate the current 365-day window and their sources have not been independently re-checked.
- Add a CI check that validates every
benchmarks.mdrow against the six-field schema (Metric, Value, Tag, Source, Date, Methodology) and a valid evidence tag. - Strengthen the thinnest sections — Structured & SQL RAG, FinOps & Cost Management, Tutorials, and Real-World Case Studies each carry only two or three entries.
Mid-term¶
- Unit tests for
scripts/discovery_engine.pycovering URL parsing and discovery markdown generation (the verified-date, benchmark-freshness, and rate-limit paths are now tested). - Extend the decision tree beyond frameworks and vector databases to embedding selection, reranking, and chunking strategy.
Longer-term¶
- A postmortem / failure case studies section, cross-linked with rag-pitfalls.md.
- Evaluate splitting the largest README sections (for example, Agentic RAG and Multimodal RAG) into side documents as they grow.
Out of Scope¶
These are deliberate decisions, not gaps: