Guardat en:
| Autor principal: | |
|---|---|
| Format: | Recurso digital |
| Idioma: | |
| Publicat: |
Zenodo
2026
|
| Matèries: | |
| Accés en línia: | https://doi.org/10.5281/zenodo.19770366 |
| Etiquetes: |
Afegir etiqueta
Sense etiquetes, Sigues el primer a etiquetar aquest registre!
|
Taula de continguts:
- <p><strong>MSP v0.3.2: A Blind Mechanism-Separation Probe for Delay, Relay, Optimization Override, and Post-hoc Legitimation</strong></p> <p>This record contains a compact methods note for <strong>MSP v0.3.2</strong>, a bounded blind adjudication scaffold designed to separate mechanism families behind directional failure after a valid update. Rather than asking only whether an updated direction is preserved or lost, MSP asks which controlled pressure regime best explains the later shift.</p> <p>The protocol integrates several earlier probe lines into a single comparative matrix and distinguishes among:</p> <ul> <li><strong>delay-sensitive erosion</strong></li> <li><strong>relay-sensitive erosion</strong></li> <li><strong>optimization override</strong></li> <li><strong>post-hoc legitimation</strong></li> <li><strong>mixed or uncertain multi-mechanism cases</strong></li> <li><strong>probe invalidity / contamination</strong></li> </ul> <p>MSP uses a shared staged backbone in which a valid update is introduced and then exposed to different controlled stresses such as delay, relay transformation, local optimization pressure, or later justificatory framing. The protocol is scored through explicit fields and bounded verdict classes under blind conditions. Across three internal blind rounds, the <strong>verdict layer remained stable</strong>, while remaining disagreement was concentrated in a small number of secondary field boundaries rather than in the top-line mechanism classes.</p> <p>The contribution is methodological. MSP does <strong>not</strong> claim to identify hidden internal causes in a strong architectural sense, nor to establish broad external generality across arbitrary models or deployments. Its claim is narrower and more useful at the present stage: within a controlled prompt regime, mechanism-sensitive directional failure modes can be <strong>operationalized, separated, and audited</strong> under blind scoring.</p> <p>This record includes the methods note associated with DOI <strong>10.5281/zenodo.19770366</strong>.</p>