Roadmap

The roadmap is intentionally short and is not tied to version numbers. Correctness remains a release requirement; the main open target is now performance.

Main priority: performance

• Use the performance benchmarks as the baseline for optimization work and keep every result reproducible through the benchmark workflow.

• Reduce fitting time, especially for the default family search. The practical goal is to close the gap to rvinecopulib and, where possible, match or outperform it without weakening correctness or API clarity.

• Profile before optimizing. Performance changes should show a measurable improvement in the benchmark suite while the correctness gate remains green.

• Prioritize pair-family selection, repeated likelihood and h-function calculations, Student-t bottlenecks, allocations, and reusable intermediate state.

• Explore parallel family or edge fitting only where benchmarks show a clear benefit.

Performance work is a particularly useful contribution area. Profiling results, lower-allocation kernels, optimizer improvements, and focused pull requests with before/after benchmark numbers are welcome.

Near-term improvements

• Improve fitted-model summaries and diagnostics.

• Reduce allocations in repeated vine evaluation and conditional-state propagation.

• Reuse or fuse intermediate computations when density and h-functions require the same quantities.

• Add data-driven truncation-depth selection with a clear statistical criterion.

• Keep fitting controls expressive without making the default API heavier.

Later

• Observation weights and deliberate missing-data handling.

• Additional nonparametric pair-copula fitting routes.

• Carefully scoped support for non-simplified vine models.