Structured Data
Release Metadata
| Version | v1.14 |
|---|---|
| Release Era | Platform |
| New Capabilities | 4 |
| Cumulative Capabilities | 58 |
| Cumulative Tests | — |
Era Context
Platform Era
GSD integration layer, planning dashboard, live metrics, session lifecycle, and promotion pipeline.
Progression Metrics
| Previous (v1.13) | 54 capabilities |
|---|---|
| This Release | +4 capabilities |
| Total (v1.14) | 58 capabilities |
| Promotion Score | 0.40/0.35/0.25 weights |
Capabilities Introduced
Architectural Decisions
- Promotion formula: determinism×0.40 + frequency×0.35 + savings×0.25 — weighted scoring reflects that determinism is the strongest predictor of safe script promotion; F1≥0.85 AND MCC≥0.75 gates prevent false positives
- Non-deterministic patterns algebraically excluded — without the determinism component (0.40), maximum achievable score is 0.60, which falls below the 0.70 promotion threshold; safety guaranteed by formula design
- Bidirectional lineage tracking — provenance queries work in both directions: from observation → pattern → promotion → script, and from script → back to originating observations
- Auto-demotion on drift detection — promoted skills automatically demoted when behavior diverges from observed patterns, preventing silent regression in the promoted script library
- SHA-256 hashes on execution captures — cryptographic integrity for observation data prevents tampering with the training signal for promotion scoring
Breaking Changes
None.