Isolated Operational Candidates / V62
Nine compact child candidates failed development acquisition gates on verified interval-cancellation tasks. The 512-case sealed and 160-case retention sets stayed locked and unexecuted. Existing weights and active aliases were preserved; V63 later tests factorization and a larger fitting budget.
Method
Following V61.2’s failed generalization and retention, nine independently owned child candidates learn from actual program outcomes using query-relative interval features. They preserve existing models. Data include 512 training, 128 development, 128 calibration, and 64 older-development cases, with 512 sealed and 160 retention cases held closed.
Results
All nine candidates failed development acquisition gates and were rejected. Sealed and retention scoring were not executed. Existing weights and active aliases remained unchanged.
Interpretation and limitations
This is failed acquisition under the recorded representation and fitting budgets, not proof that the architecture cannot learn the task. V63 later changes factorization and budget and compares a stronger flat control. Ownership isolation preserves the baseline but does not itself establish capability acquisition.
SOURCE PROVENANCE
EMMA V62: compact verification acquisition and preserved routes
LABORATORY REPORT / 2026-10-06SOURCE CHECKSUM / SHA-256
8c2caa396faeee0d94f39877fa1f753acbaca3845d4b9290c753a5d57e9d1e5ePublic journal edition reviewed 2026-10-08. 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.