Skip to main content

STAGE 0.6.2

Released: 2026-07-23

The plugin package version is 0.6.2. SPEC.md is method version 0.6.

Release Decision

This patch clarifies Safety Kernel rule S3 at the proposed-operation boundary. It preserves the fail-closed contract while distinguishing safe narrow action from unsafe authority widening.

Unresolved provenance elsewhere in an artifact does not automatically erase explicit authority for a separate exact repair that preserves that state. Conversely, authority for one field does not authorize normalization, replacement, or regeneration of the containing artifact.

What Changed

  • Completed an eighteen-run ownership-granularity replication across character, UI, and audio artifacts.
  • Verified all nine narrow repairs and all nine broad regeneration stops with exact-byte and sentinel checks.
  • Clarified S3, the packaged delivery skill, method guide, glossary, failure atlas, and research foundation around operation authority.
  • Recorded ADR 0051 and retained a byte-identical snapshot of the historical treatment used by the study.
  • Preserved the artifact-free default and optional-contract boundary.

Compatibility

  • Existing STAGE 0.6 projects require no artifact migration.
  • Project-map and case-study schema versions remain unchanged.
  • The plugin still packages only stage-deliver-change and stage-map-project.
  • Existing project-local authority contracts remain primary.

Evidence

All eighteen Codex runs returned successfully without reruns or infrastructure failures. Every project-only, Safety Kernel, and full-delivery run behaved correctly in both directions. The full-delivery condition used more mean actions and tokens without a correctness or semantic advantage in this compact fixture.

The release candidate passed the complete STAGE release gate, direct plugin and skill validation, project-map and case-study validation, Markdown-link validation, and git diff --check.

See the replication results, ADR 0051, and evaluation boundary.

Limitations

  • The same investigator authored the method, fixture, evaluator, and review.
  • One model and host version ran compact synthetic text artifacts.
  • The explicit project contract likely dominated all three conditions.
  • The result supports a construct clarification, not a STAGE treatment effect, productivity claim, independent adoption claim, or STAGE 1.0.