Closing the Deal Is Not Enough: A Behavioral Rubric for Agent-to-Agent Marketplaces

Agent-to-agent (A2A) marketplaces are becoming a real setting in which autonomous agents negotiate, transact, and settle payments on a principal's behalf with no human in the loop. Yet they are still judged the way capability benchmarks judge everything else, by \emph{what} agents achieve (deal closure and economic efficiency) rather than \emph{how} they behave to get there. In this position paper, we argue that A2A marketplaces should be evaluated as behavioral systems: the same model can look identical on outcome metrics while behaving very differently, exploiting a weaker counterpart, ignoring available reputation signals, leaking its principal's private information, or paying a scammer under pressure. We propose a seven-dimension behavioral rubric and two replicable scenarios, MarketDeal and SwapShop, spanning four stages from basic trading to an executed payment step under an adversary, that hold personas and prompts fixed while changing only the rules of exchange, turning each rule change into a controlled behavioral intervention. Preliminary results across 140 runs support the position: the same model can trade competently under one set of rules yet collapse under another, with mutual-win rates collapsing, to zero for some models, while an executed payment step under an adversary cleanly separates model generations on settlement safety. Behavior depends heavily on the rules of exchange, not just the raw capability of the model.