docs: attribute the reported task-success result to the dated paper
#4
by CyberNativeAI - opened
The dataset card currently calls 34.1% the best-model success rate in two places. Its cited paper, arXiv:2603.13594v1, reports 37.4% for Claude Opus 4.5 in Table 2. This change attributes that dated paper result and removes the repeated unqualified best-model assertion. It does not claim a current-model ranking or change dataset metadata, loading instructions or task counts.
Validation used static parsing of all 16 configuration rows in the cited table and checked that only two prose lines change. No models, benchmark environment or dataset tasks were executed; the paper's measurements were not reproduced.
Prepared with AI assistance by CyberNative AI LLC. Correction contact: hello@cybernative.ai.