HarnessEval-W

HarnessEval-W is an evaluation product applying the harness paradigm to world models, developed by Writer. It uses specialized sub-agents to generate auditable reasoning chains behind every score, enhancing transparency in AI model assessment.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.