Evaluating Generative Agents with Actions Grounded in Socially Distributed Task Environments using Incognita

Effective agency in social environments depends on when an agent seeks knowledge, when it acts, and whether its actions are justified by the information it has acquired. Existing grounded benchmarks provide executable actions, persistent state, and verifiable outcomes, while social simulation environments provide rich interaction among language agents. We study an evaluation setting that brings these requirements into a single environment. We define socially distributed task environments as interactive environments in which task-relevant knowledge is partitioned across role-isolated participants, and consequential actions are reachable only through interaction with those participants. Under this view, communication functions as exploration over role-partitioned knowledge, and grounded action functions as exploitation over consequential environment state. We introduce Incognita, a Concordia-based environment framework that instantiates this setting. Incognita separates social interaction from grounded execution. The evaluated agent routes messages to a user or specialist entities; specialist entities mediate admissible operations; a deterministic sub-environment executes accepted operations over a canonical state; and an offline evaluator scores final outcomes using inherited task rewards. Incognita-Retail transforms tau-bench retail into a multi-entity environment while preserving its final-state reward semantics. We evaluate three generative agent models on 18 tasks stratified by social breadth, with 540 total trials. Model progress appears in both reward and behavior. Success rises from 0 percent to 8.9 percent and 17.2 percent, while premature finalization falls from 100 percent to 87 percent and 58 percent. Stronger models elicit more hidden knowledge, contact more entities, and attempt more grounded writes, yet reliability remains low. These findings suggest that socially distributed task environments expose agent behavior before reliable task success, including knowledge elicitation, source selection, grounded action attempts, and premature completion belief.