← back to paper
arxiv: 2607.09328 · 2 revisions
WildTrace: Benchmarking Natural Evidence Trails in Long-Context Reasoning