15.231 14.8.21 Perfiles de benchmark y claims
| Perfil | Claim permitido | Etapas requeridas |
|---|---|---|
BP0-SANITY |
pipeline executes and beats trivial baselines | VAL-STATIC VAL-SYNTH |
BP1-COMPONENT |
component research result | VAL-COMP |
BP2-INTEGRATED-LAB |
integrated replay in controlled gym capture | VAL-INTEGRATION VAL-SYSTEM |
BP3-EXTERNAL |
generalization beyond development domains | VAL-SHIFT |
BP4-OPEN-WORLD |
safe unknown/abstention behavior | VAL-OPEN |
BP5-EDGE |
target hardware capability at quality floor | VAL-SYSTEM VAL-SHIFT |
BP6-FIELD-PILOT |
prospective operational pilot | VAL-FIELD |
BP7-COACH-UTILITY |
decision-support utility | VAL-HUMAN |
Un resultado BP1-COMPONENT no autoriza un claim de
despliegue. Un resultado BP5-EDGE no demuestra utilidad
para entrenadores. Los claims sólo ascienden al completar las etapas
correspondientes.