Relay 0.2.4 retains the semantic boundary established by the 0.2 series and
the versioned schemas introduced in 0.2.1. The current release includes
qualified identities, async-aware Python handling, conservative
cross-language profiles, explicit model metadata, reachability-scoped impacts,
resource bounds, atomic reports, and reproducible-build verification.
Version 0.2.4 makes relay doctor fail on missing paths and points schema $id URLs at the v0.2.4 tag; it does not add analyser rules or broaden those models.
For 0.3.0, I am thinking about:
- add source-level suppression directives with auditable reasons;
- add opt-in compiler, Tree-sitter, or language-server front ends behind the existing capability boundary;
- measure per-rule precision, recall, and false-positive budgets before marking any detector stable;
- add mutation tests for rule boundaries and corpus regressions;
- expand execution-context contracts without implying scheduler simulation.
Each item is tracked in GitHub issues. The order may change when a reproducible example shows that another problem matters more; that is one of the nice parts of growing this as an open personal project.