Combining Two Learning Mechanisms / V108.2
Combining continuous correction with self-found scoped improvements reached 84.5% on 3,000 sealed configurations, against 80.7% for the core and 84.1% for continuous correction alone. The +0.4-point gain over the best single mechanism missed the +1-point gate: at this volume the two mostly substitute for each other.
SCOPE + LIMITATIONS
- One task family and one stream. By gain per cost continuous correction is preferred; scoped parts keep their value where a need calls for an inspectable part that must stay unchanged outside its scope.
- This public edition reviews saved evidence; experiments were not rerun for publication.
CONTROL COMPARISON
From one stream of 6,000 verified outcomes, build five pathways: the core alone, each mechanism alone, and both in each order. Score each once on 3,000 sealed configurations across six regimes.
Failed the complementarity gate (+0.4 against +1); passed +3.8 points over the core and no regime below it. No complementarity claim is made, and the demonstration keeps the better order. The run also showed the system meeting a need by creation without training once the stream was large enough.
- Core
- 80.7%
- Continuous correction
- 84.1%
- Scoped improvements
- 83.5%
- Scoped, then continuous
- 84.5%
- Continuous, then scoped
- 84.1%; no further need found
- Need met by creating a new part without training
- Found and attached for the first time
Question and method
From one stream of 6,000 verified outcomes, build five pathways: the core alone, each mechanism alone, and both in each order. Score each once on 3,000 sealed configurations across six regimes.
Recorded results
| Condition | Recorded result |
|---|---|
| Core | 80.7% |
| Continuous correction | 84.1% |
| Scoped improvements | 83.5% |
| Scoped, then continuous | 84.5% |
| Continuous, then scoped | 84.1%; no further need found |
| Need met by creating a new part without training | Found and attached for the first time |
Comparative standing and locked gates
Failed the complementarity gate (+0.4 against +1); passed +3.8 points over the core and no regime below it. No complementarity claim is made, and the demonstration keeps the better order. The run also showed the system meeting a need by creation without training once the stream was large enough.
Interpretation and limitations
One task family and one stream. By gain per cost continuous correction is preferred; scoped parts keep their value where a need calls for an inspectable part that must stay unchanged outside its scope.
SOURCE PROVENANCE
Combining Two Learning Mechanisms / V108.2
LABORATORY REPORT / 2026-10-26SOURCE CHECKSUM / SHA-256
092135224078a1ccef25a702f4f5f3e3bafdb5945b785bb7660d5f327ceca0c5Public journal edition reviewed 2026-10-28. Source documents and saved evidence were inspected; experiments were not rerun for this edition. Proprietary implementation code, model binaries, private infrastructure, and detailed machine records are not published here. Journal identifiers are editorial references. Catalog inclusion does not imply qualification or runtime promotion.