This roadmap is a release-state view, not a historical task board. It separates what is in the 0.9.2 artifact line from work that exists only on main and from forward-looking architecture priorities.

​
Released Baseline: 0.9.2

The 0.9.2 line is the current public release baseline. It includes the shared Rust workspace, CUDA-backed deterministic execution, probabilistic and epistemic surfaces, solver services, Python packaging, and the xlog CLI. The public Rust release boundary is the set of publishable crates listed in Release Process. pyxlog is shipped through Python packaging, not crates.io.

​
Stable Architecture Commitments

XLOG’s architecture is organized around these commitments:
  • one frontend for deterministic, probabilistic, epistemic, and solver-backed programs;
  • explicit host control versus GPU data-plane boundaries;
  • RIR/EIR/PIR/SIR artifacts that make subsystem contracts inspectable;
  • CUDA kernels that operate over device-resident buffers;
  • fail-closed behavior when a proof, route, or capacity contract is not met;
  • route counters and telemetry for optimized execution paths.
These are the commitments public docs should emphasize over old milestone labels or implementation-board names.

​
Main-Only Work Beyond 0.9.2

The repository main branch contains implemented work that is not part of the 0.9.2 release artifacts:
  • aggregate-fused WCOJ for selected triangle, 4-cycle, and clique aggregate shapes;
  • GPU Free Join for general multiway bodies and factorized count-by-root;
  • dense and sparse factorized recursive deltas;
  • factorized non-count aggregate folding for probabilistic provenance;
  • joint multi-rule neural-symbolic mixtures;
  • Stage-B existential joins;
  • grouped forward/backward training paths.
Document these features as “on main, unreleased beyond 0.9.2” until a new public release includes them.

​
Near-Term Documentation Priorities

The docs roadmap is now:
  • keep docs-site/ as the public source of truth for xlog.md;
  • keep local agent reports, plans, evidence, and superpower notes out of Git;
  • replace internal milestone prose with source-backed product documentation;
  • add accurate diagrams that show host/device and IR boundaries;
  • keep generated static HTML in the docs-dist deployment branch, not in source docs.

​
Architecture Priorities

The next architecture work should stay focused on runtime facts that users can observe:
  • clearer route telemetry for WCOJ, Free Join, aggregate fusion, and recursive delta routing;
  • stronger release evidence for CUDA-required behavior;
  • explicit no-host-transfer counters where a path claims device-resident data flow;
  • sharper failure diagnostics for unsupported probabilistic, epistemic, and solver shapes;
  • docs examples that show how to verify route dispatch instead of inferring it from final answers.

​
Out-Of-Scope For This Roadmap

This page intentionally does not preserve the old line-by-line historical roadmap. Internal phase labels, board codes, and dated task bundles are not public product documentation. When a historical item matters to users, document the shipped behavior, the command that verifies it, and the release boundary.