Skip to content

ACE Core 0.6.0 — Measured Intelligence

Choose a tag to compare

View all tags
@eamirian eamirian released this 11 Aug 20:54
· 74 commits to main since this release
v0.6.0
1e383e1
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

ACE Core 0.6.0 — Measured Intelligence

ACE Core 0.6.0 adds the first durable, domain-neutral measured-impact slice. Under explicit product-owned criteria, ACE can classify a governed intelligence artifact or cognition revision as useful, harmful, or unproven, while preserving exact attribution, uncertainty, controls, provenance, and append-only replay.

Highlights

  • Versioned criteria, conditions, evidence, evaluation, and governance-proposal contracts link an exact evaluand to material use, Decision, reviewed Action, observed result, Outcome, controls, and cutoff provenance.
  • Missing attribution, mismatched conditions, future leakage, unavailable outcomes, duplicate evidence, and insufficient support fail closed or remain explicitly unproven.
  • Evaluation and proposal receipts are appended atomically and replay exactly across restart; divergent replay and scope drift are rejected.
  • Promotion, rejection, rollback, and retirement remain proposals only. They are non-effective, non-selectable, and never silently applied.
  • Owner-governed Monitor and Subscription lifecycle contracts support durable create, pause, resume, revoke, and explicitly requested sensing windows without granting scheduler, source-transport, delivery, or external-action authority.
  • The separately packaged reference workspace action adapter advances to distribution version 0.2.0 and requires ace-core>=0.6.0,<0.7; its capability implementation identity remains 0.1.0.

Install Core with:

pip install ace-core==0.6.0

The reference adapter wheel and source archive are attached to this GitHub Release rather than published as part of the Core distribution.

Boundary and non-claims

This release proves a bounded matched-evidence mechanism, not general causality, universal benefit, self-certification, automatic promotion, distributed exactly-once effects, or untrusted-code isolation. Issue #49 F3 remains open under its documented containment with a 2026-11-05 deadline. The World Intelligence public-artifact reproduction is a separate external release gate and will be recorded independently.

See the repository changelog and Measured Intelligence work-packet/evidence records for the exact supported boundary.

Assets 4
Loading