MAST-U feedback label collection passes local artifact checks
We completed the owner-approved feedback label collection for MAST-U remediation. The collection uses training equilibria and excludes both test equilibria. The surrogate, guard and closed-loop acceptance limit remain unchanged.
We reviewed the finished local artifacts before considering training. The review checks the allocation against the approved resource and time limits, verifies checksums and provenance, and compares every label input row with its selected case and source trajectory. It also checks the case budget, unique coordinates, training-only equilibrium selection, source-model identity and finite successful labels: eight local checks pass for approved allocation/time, hashes and lineage, full-input alignment, case budget, unique coordinates, training-only equilibria, source model and finite successful labelsC-183.
| stage | result |
|---|---|
| completed collection | eight local checks pass for approved allocation/time, hashes and lineage, full-input alignment, case budget, unique coordinates, training-only equilibria, source model and finite successful labelsC-183 |
| artifact synchronization | awaiting explicit approval |
| training and coupled revalidation | pending |
The new evidence establishes that the collected labels retain their full solver inputs and their source lineage. It does not establish held-out model accuracy or an improved temperature trajectory. The existing failed MAST-U acceptance result remains open (K-030).
What’s next. Synchronize the reviewed artifacts after approval, then prepare training and validation while preserving the test split and the owner-required monotonicity constraint. Another coupled comparison must assess the unchanged closed-loop limit before any claim of successful remediation.
Technical details → Verification › mastu remediation
Decisions → OQ-27