A new arXiv paper introduces TxnQA, a benchmark designed to test analytical reasoning over merchant transaction event sequences. The authors argue that analyzing such histories requires dynamic temporal filtering, entity resolution, and aggregation over irregular event streams—operations that standard static tabular or time-series forecasting benchmarks do not capture well.
TxnQA is positioned as a way to evaluate how well models can ground their reasoning in evidence from transaction data. By focusing on the messy, event-driven nature of real merchant records, the benchmark aims to expose limitations in current approaches and push toward more robust analytical capabilities.
The paper's contribution is primarily the benchmark itself, along with the framing of transaction-sequence analysis as a distinct challenge. Since the abstract is brief, the full methodology and results are not detailed here, but the motivation is clear: existing evaluation suites miss the kind of reasoning that transaction analytics demands.