RuleChef roadmap: trustworthy routing and richer local-rule research

[adaamko]

We have published a deliberately lean, evidence-driven RuleChef roadmap.

The near-term v0.3 focus is correctness that remains useful regardless of future research:

• split before synthesis and preserve evaluation roles;
• support- and confidence-aware fallback;
• fail closed on schema-invalid local output;
• correct transformation and routing metrics;
• consistent version metadata.

Separately, the undated richer-local-rules research milestone is a collaborator playground for spaCy, semantic/vector prototypes, and a minimal RuleIR design. Those are experiments and RFCs, not committed stable APIs.

Milestones:

• v0.3 — Trustworthy Learning and Routing
• Research — Richer Local Rules

Feedback is especially welcome on whether this separation is clear and which current correctness problem matters most in real use.