Skip to content

0.7.26: Evidence-Driven Actor Efficiency

Choose a tag to compare

@llblab llblab released this 13 Sep 03:04
· 1 commit to main since this release

Highlights

  • Retains the placed pair-owner fallback for Crossing admission, improving the measured steady cycle from 12 to 16 actors at unchanged cycle ProofSize while preserving bounded scalar fallback behavior.
  • Adds and binds the generated Crossing admission-selection Weight owner, then reissues the production runtime Wasm, metadata, and generated consumers.
  • Records the complete Actor resource-anatomy study and closes the frozen six-candidate redesign campaign without adopting an unsafe or unproven optimization.
  • Confirms the converged runtime across ten end-to-end workloads with unchanged counts, per-block values, proof totals, and state effects.
  • Restricts placed pair fallback to tail-page authority so constrained non-tail batches preserve scalar progress rather than halting Crossing service.
  • Reconciles Actors integration, package architecture, and bilingual Wiki documentation to the measured implementation truth.