Perplexity Open-Sources WANDR — a 500-Task Research Benchmark Built on Production Data With Dynamic Evidence Verification

Latent Space (AINews) / Perplexity·medium signal

Perplexity released WANDR, a 500-task benchmark for research agents built from production query data and using dynamic evidence verification rather than static gold answers. Grounding evaluation in real web-research traffic targets the exact failure mode of deep-research agents — plausible but unverified synthesis. For anyone building or evaluating a research/RAG agent, an open, production-grounded benchmark with verification baked in is a directly usable measuring stick.