Skip to content

YNM 1.4.0: adaptive execution and empirical evaluation - #3

Merged
RobLe3 merged 22 commits into
mainfrom
codex/ynm-1.4.0-adaptive-execution
Aug 20, 2026
Merged

RobLe3 merged 22 commits into
mainfrom
codex/ynm-1.4.0-adaptive-execution

Conversation

@RobLe3

@RobLe3 RobLe3 commented Aug 20, 2026 •

Copy link
Copy Markdown
Owner

Summary

  • adds YNM-0 through YNM-3 adaptive execution and evidence-bounded escalation
  • strengthens affirmative YES requirements and bounded absence handling
  • defines YNM as bounded evidential decision support and adds YNM-BRP-1
  • preserves all frozen empirical findings and adverse results
  • prepares 1.4.0 for publication as a bounded Research/Experimental Pre-release
  • adds a direct public explanation, practical example, measured benefits and costs, and an accessible workflow diagram

Validation

  • repository/schema validation: PASS
  • release-integrity validation: PASS
  • deterministic tests: 98 PASS
  • package build and deterministic content signature: PASS (0552f68ea5129c0a1f4decc41a50fdef60b50a0b0c1f28cac9e2c85b9e3754af)
  • Agent Skills reference validation and properties: PASS
  • security-boundary validation: PASS
  • public sanitization: PASS

Research evidence

The primary 1.4 holdout showed full material recall, 0.9722 precision, no unsupported claims, full required-MAYBE recall, complete evidence traceability, and no authority violations. PORTABLE execution also used approximately 2.34× input tokens, 2.82× output tokens, 2.40× elapsed time, and 2.90× tool calls relative to control.

Cross-model non-inferiority failed under the frozen criteria. YNM-BRP-1 also remained NO after missing one frozen required-MAYBE criterion, despite strong usefulness, traceability, inspectability, and containment. Activation and acceleration remain MAYBE; independent interoperability remains MAYBE.

No fixture, threshold, adjudication, finding, or historical disposition was changed or reclassified.

Publication decision

Version: 1.4.0
Release class: Research / Experimental Pre-release
Development state: paused after the completed research cycles

The maintainer has authorized publication under the narrower claim that YNM is usable as bounded, auditable advisory review with disclosed limitations. This does not claim universal effectiveness, production certification, cross-model equivalence, or cost reduction.

After this PR passes CI, the intended sequence is merge commit → human-authorized finalization on main → v1.4.0 tag → tag-integrity CI → GitHub Pre-release.

@RobLe3
RobLe3 merged commit d8b944f into main Aug 20, 2026
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant