Evidence-gated development

Roadmap to stable Core-1

The roadmap is ordered by semantic risk, not feature count. A pillar is not complete until source, checked representation, independent execution, failures, cleanup, and cross-domain boundaries agree.

01
Working now

Algorithm foundation

A bounded Core-0 slice runs through the checked interpreter, portable C materializer, LLVM JIT, and AOT. Richer ownership and cleanup remain explicit gaps.

  • Versioned ZLM1 artifacts
  • Structured diagnostics
  • Cross-engine differential tests
02
Preview integration

Five-domain Core-1

Domain-aware semantic identities and ZLM2 regions now support bounded source-backed compute, state, flow, and optimize work. Each preview is still narrower than its normative authority.

  • Typed domain boundaries
  • Reference and independent evaluators
  • Fail-closed unsupported regions
03
Current focus

Vertical conformance

Replace broad runtime prototypes and benchmark scaffolds with end-to-end source constructs, independent oracles, durable evidence, and complete package-closure adaptation.

  • Task-specific ZL256 fixtures
  • Signed public evidence
  • Interpreted ↔ compiled closure validation
04
Stable gate

Core-1 activation

Stable Core-1 activates only after all five domains have independent materialized execution, canonical identities and boundaries, persistent-flow recovery, and mutation-resistant conformance.

  • Apple Silicon SIMD CI
  • Kill/restart flow tests
  • Core-0 and ZLM1 compatibility
Self-hosting is deliberately off the critical path.

Compiler provenance is useful evidence, but it does not replace language semantics, independent execution, or public conformance.