Qualified Opening-Step Stem / V75.3
A default tree-ensemble module reached 97.62% on 800 sealed runs after observing 120 of 200 steps, versus 85.12% from configuration alone. Its locked gates passed and it was attached to the prototype. Only the 120-step module is qualified; shorter-opening scores are measurements.
CONTROL COMPARISON
Fit an owned gradient-boosting module for each opening length on 3,600 training runs. Development uses 400 newly executed runs; sealed evaluation opens the 800 configurations left untouched by earlier V75 versions. Only the 120-step module is designated for qualification.
The gated 120-step module passed development, sealed, calibration, preservation, executor-parity, and fresh-process restoration checks. It improves on configuration-only prediction, but uses a standard boosting method and more information; no novel learning-method advantage is claimed. Shorter rows were not gated.
- Configuration only · sealed
- 85.12% accuracy; 0% execution
- Forty steps · ungated
- 92.50%; 20% execution
- Sixty steps · ungated
- 94.38%; 30% execution
- Eighty steps · ungated
- 95.75%; 40% execution
- 120 steps · qualified
- 97.62%; 60% execution; calibration error 0.012
Question and method
Fit an owned gradient-boosting module for each opening length on 3,600 training runs. Development uses 400 newly executed runs; sealed evaluation opens the 800 configurations left untouched by earlier V75 versions. Only the 120-step module is designated for qualification.
Recorded results
| Condition | Recorded result |
|---|---|
| Configuration only · sealed | 85.12% accuracy; 0% execution |
| Forty steps · ungated | 92.50%; 20% execution |
| Sixty steps · ungated | 94.38%; 30% execution |
| Eighty steps · ungated | 95.75%; 40% execution |
| 120 steps · qualified | 97.62%; 60% execution; calibration error 0.012 |
Comparative standing and locked gates
The gated 120-step module passed development, sealed, calibration, preservation, executor-parity, and fresh-process restoration checks. It improves on configuration-only prediction, but uses a standard boosting method and more information; no novel learning-method advantage is claimed. Shorter rows were not gated.
Interpretation and limitations
Small synthetic runs only. The module costs sixty percent of full execution and is attached as a prototype stem. A graded escalation policy and its average cost remain untested. The later branching audit estimates what additional initial-state information might recover.
SOURCE PROVENANCE
EMMA V75.3: the opening-steps judgment at longer openings
LABORATORY REPORT / 2026-10-08SOURCE CHECKSUM / SHA-256
5f9e61cbe7abb9c74191b9494ff0038c8190cb3469c83b65c466d123201d5951Public 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.