Operation Registration by Description / V74.4
Adding operation descriptions did not establish open registration for the frozen V74.3 head. It selected new operations for 7/20 and 8/20 requests versus zero-shot Laya’s 19/20 and 18/20. All three fixed criteria failed; no module was changed.
CONTROL COMPARISON
Load the V74.3 head read-only without its decline threshold and offer six operation descriptions. Score 100 requests, including forty for two new operations, forty existing answerable questions, and twenty unsupported requests. This is a probe with no fitting.
All three registration criteria failed: at least 80% on new operations, at most one existing answer lost, and at most one control capture. Laya is stronger at matching new descriptions but declines none. The subsequent fine-tuned Laya probe in the V74 report also loses open matching, preventing a claim that either trained system solves registration.
- New field-list operation · EMMA / Laya
- 7/20 / 19/20
- New duration operation · EMMA / Laya
- 8/20 / 18/20
- Existing answerable questions retained
- 32/40 / 40/40
- Unrelated control-description captures
- 21/100 / 0/100
- Unsupported requests declined
- 12/20 / 0/20
Question and method
Load the V74.3 head read-only without its decline threshold and offer six operation descriptions. Score 100 requests, including forty for two new operations, forty existing answerable questions, and twenty unsupported requests. This is a probe with no fitting.
Recorded results
| Condition | Recorded result |
|---|---|
| New field-list operation · EMMA / Laya | 7/20 / 19/20 |
| New duration operation · EMMA / Laya | 8/20 / 18/20 |
| Existing answerable questions retained | 32/40 / 40/40 |
| Unrelated control-description captures | 21/100 / 0/100 |
| Unsupported requests declined | 12/20 / 0/20 |
Comparative standing and locked gates
All three registration criteria failed: at least 80% on new operations, at most one existing answer lost, and at most one control capture. Laya is stronger at matching new descriptions but declines none. The subsequent fine-tuned Laya probe in the V74 report also loses open matching, preventing a claim that either trained system solves registration.
Interpretation and limitations
Two new operations and one teacher request draw, with one description per operation. Existing independent questions are now development data. Operations supplied as text are not yet shown to function as open data for the EMMA Decision.
SOURCE PROVENANCE
V74.4: registering an operation by description alone
LABORATORY REPORT / 2026-10-08SOURCE CHECKSUM / SHA-256
020a7714836714753953965fba326e8bf6e5f194882d57f5217cdbbadfa00cedPublic journal edition reviewed 2026-10-09. 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.