Answers are grounded in this paper and cite their sources.
Independent test reveals value headroom moderates negotiation success in language model agents, suggesting critical implications for AI interactions.
When two language-model agents negotiate, does it matter more how they are styled (warm, dominant) or how their objective is specified (ranked priorities, a reservation value, a concession rule)? A large autonomous-negotiation study reported that interpersonal warmth, not structure, predicted success. This note offers an independent test – a reconstruction from that study’s public materials, not a one-to-one reproduction – and identifies value headroom (the joint surplus a naively cooperative pair leaves on the table) as a first-order moderator. Across seven arms spanning three model families (4,920 dyads), specification-first prompting ties interpersonal style on near-ceiling scenarios (a preregistered null) but yields strictly higher joint value once headroom is present (Cohen d = .314, p = .009 on a mid-capability model; d = .569, p < .001 at the frontier – the advantage grows with capability), with a graded design confirming the headroom slope. An ablation shows the advantage comes from teaching logrolling, not leaking the payoff matrix; it survives prompt paraphrase and replicates across model families. Specification moves the objective value frontier while remaining affectively neutral; style games the agreement proxy. The apparent “style beats structure” result is recovered as the zero-headroom boundary case of a more general law. Includes zharnikov-2026aq-negotiation-spec-experiment.yaml (Paper Spec v0.1.0) – a machine-readable specification of the paper's claims, assumptions, and dependencies. The paper's full machine-first bundle (the SPINE claim/dependency graph and the ONTOLOGY term module) lives in the public repository; see https://github.com/spectralbranding/paper-spec for the standard. This PDF is generated programmatically from that machine-first source under a research-as-repository model.
No takes yet. Share an insight, caveat, or question.
Dmitry Zharnikov (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: