Scale AI

Data-and-evaluation company for AI systems; publishes research under Scale AI Research. In this spoke it enters as the sourceOrganization of model-or-harness-taxonomy (arXiv:2607.28802), the interaction-centric taxonomy behind agent-failure-localization, authored by Harsh Raj, Vipul Gupta, Anas Mahmoud, Razvan-Gabriel Dumitru, Darvin Yi, Aakash Sabharwal and Yunzhong He (@scale.com).

Thin node — the evidence held here is the one paper. It is worth noting what kind of vendor is publishing it: an evaluation-and-data company producing a scheme for attributing agent failures sits adjacent to its commercial interest, and the taxonomy is validated on the authors' own 40 annotated examples. That does not make the work weak (the agreement statistics are published and the limitations stated), but it is not the neutral third-party evaluation the spoke's growth edge #3 asks for.