Synthetic dialogue corpora are increasingly used as proxies for target data, yet persona-grounded generators optimize individual conversations rather than corpus composition, yielding plausible dialogues with distorted population-level behavior mixes. We introduce GroupPersona, a framework that aligns synthetic corpora with a reference behavior distribution. GroupPersona turns population statistics into generation controls by discovering recurring behavioral groups in the reference corpus and using their prevalence and interaction patterns to condition user agents.

We evaluate corpus-level fidelity through behavioral alignment and structural preservation, measured by average Jensen–Shannon distance over 12 behavior attributes and 5 dialogue structure attributes, respectively. Across four corpora spanning two source families and two construction regimes, GroupPersona reduces behavioral Jensen–Shannon distance from 0.234 to 0.177 over the best-average baseline, a 24.4% reduction. It is best or tied-best on all four corpora. It maintains comparable structural distance at 0.167 versus 0.169. Together, these results show that GroupPersona improves population-level behavioral fidelity while preserving dialogue structure.

Agentic task-oriented dialogue (TOD) requires systems to track concurrent goals, dependencies, and long-horizon state. We examine goal-lifecycle recovery from fixed dialogue trajectories. ATOD contains 1,000 synthetic dialogues annotated for six advanced-TOD properties, and ATOD-Eval defines metrics for dependency-sensitive completion, memory recall, and proactivity. We implement a symbolic–vector memory evaluator for ATOD-Eval. Among five prompt-only predictors and six matched-backbone memory baselines, our evaluator is the only configuration above 90% in both goal detection F1 and conditional status accuracy on medium dialogues. On complex dialogues, no baseline exceeds it on both metrics; it has the highest conditional status accuracy within the memory-based block and the lowest measured per-turn latency. These experiments assess lifecycle tracking rather than interactive agent task success. A cross-family judge swap and a manual audit with 94.0% agreement provide initial checks on measurement reliability. Code and data will be released at https://github.com/amazon-science/ATOD.