Enterprise AI assistants suffer from two failure modes not captured by existing public-web QA benchmarks: authorized information leakage, where factually correct answers reach unintended readers, and aggregation failure, where evidence scattered across artifacts yields fluent but incomplete responses. Both stem from enterprise authority being per-item and partly body-embedded, a structure that existing benchmarks address along at most one axis. We argue that authority-aware evaluation must be conducted at the per-item level, and instantiate this principle through Cross-Artifact and Authority-aware Reasoning over Organizational Bundles (CAROB): a synthetic enterprise dataset in which every retrieval-and-citation item carries a time-indexed audience/approval/lifecycle annotation, paired with three tasks that decouple cross-artifact coherence, authority-aware retrieval, and role-conditioned answering. Evaluating five frontier and open-weight LLMs, we find that no single model dominates both axes of the benchmark: each task induces a distinct model ordering, and the model strongest on authority handling ranks last on answer coverage. Two structural patterns hold across the lineup: retrieval-side leakage amplifies substantially at the answer stage, and structural authority filtering eliminates the residual leak band that prompt-level mitigation leaves behind, with both patterns replicating on EnronQA and ACCESS DENIED INC. We release CAROB as an evaluation testbed for authority-aware enterprise system design, together with its per-item annotation, three-task structure, and label-first generation pipeline as reusable infrastructure.

Tables in real-world documents, including scientific papers, financial reports, and statistical summaries, frequently use merged cells, multi-level headers, and sectioned layouts. Yet the lightweight formats LLMs handle best (Markdown, CSV) cannot express these structures, while expressive formats (HTML, LaTeX) impose excessive token cost and parsing complexity. To enable systematic analysis of this trade-off, we introduce a taxonomy decomposing table structure into six atomic features (F1–F6) and corresponding complexity levels (L1–L6), and construct STB (Synthetic Table Benchmark), a controlled benchmark with 1,089 tables systematically distributed across complexity levels. Using this framework, we propose LTEM (LLM-friendly Table-Enhanced Markdown), a formally specified, minimal extension of Markdown that explicitly encodes spans and hierarchical headers. Experiments across five state-of-the-art LLMs show that LTEM achieves accuracy comparable to HTML, offers a favorable accuracy–token trade-off relative to HTML and LaTeX, and produces substantially fewer invalid responses than LaTeX; for table generation, LTEM is comparable to MultiMarkdown though HTML is best. We release our taxonomy, benchmark, and format specification to support future research on LLM-table interaction.