This study focused on the experiences of researchers who had used social sim-ulation methods in the context of policy development. Participants (n = 20) were surveyed about several aspects of their projects, including 1) project de-scription and demographics, 2) aspects of model development, 3) collabora-tions involved in the project and how they influenced project outcomes, and 4) project outcomes. Participants’ modelling projects were often commissioned directly by governments (e.g., through competitive calls for research funding or through contact with policy professionals) and commonly required research-ers to have expertise that predated the policy problem. Policy models were most often developed for explanatory purposes, rather than prediction. Collab-orators’ varying levels of familiarity with social simulation approaches had minimal impacts upon projects, as teams were able to address gaps through training sessions and workshops. Most collaborators understood the model and outputs well by the end of their projects. Significant correlations were ob-served between collaborators’ trust in the data produced by the model, their understanding of policy and practice implications, and their trust in the model outputs. Despite common challenges of time pressures, model/project scale, data, and resources, interdisciplinary collaborations were also described posi-tively, as models and their insights were useful to stakeholders. These insights paint a different picture to the often-reported challenges associated with policy modelling and offer a nuanced perspective about the realities of engaging with policy-relevant work.

This extended abstract introduces the SURIMI model graphical user interface (GUI), allowing for exploring what-if simulation scenarios in the SURIMI use cases.

Agent-based models (ABMs) are powerful tools for studying

complex adaptive systems, yet the ways researchers disseminate their

results remain poorly aligned with the epistemic aims of social simu-

lation practice. Following Epstein’s question of why we model and

Edmonds’ reflections on modelling purpose, this paper reframes the

discussion around a central question: what modelling for? If modelling is

meant to expose reasoning to scrutiny, enable replication, invite exten-

sion, and build cumulative knowledge, then current practices—especially

static graphical summaries—systematically fail agent-based modelling

as a science. ABMs generate high-dimensional behavioural landscapes

shaped by stochasticity, path dependence, and parameter interactions,

but academic formats compress these landscapes into one or two static

figures. This creates information loss, hampers transparency, and dis-

courages reuse. Building on trends in standardization (ODD protocol),

open-access dissemination (CoMSES Net), and participatory modelling,

we argue for interactive, web-based publication of simulation outcomes as

a mode of knowledge dissemination that aligns with the epistemic goals

of social simulation practice. Using two Shiny applications as examples,

we demonstrate how interactive dissemination can reveal behavioural en-

velopes, allow exploration of parameter sensitivities, and mitigate cherry-

picking. We conclude that opening the black box of ABM results is a

methodological necessity for achieving cumulative social simulation sci-

ence.

Agent-based models record every interaction, opinion shift, and relationship change for every agent at every time step -- yet this individual-level richness is routinely discarded in favor of aggregate statistics. The result is a communication gap: the audiences who most need model insights process information through narrative, not parameter sweeps. Bridging this gap would let modelers reach policymakers, stakeholders, and interdisciplinary collaborators on cognitive terms that favor comprehension and action. Can individual agent trajectories be systematically converted into structured narrative without sacrificing analytical rigor?

This paper adapts digital prosopography -- the construction of collective biography from fragmentary records of individual lives -- as an output pipeline for agent-based simulation. Event logs from a bounded confidence opinion dynamics model on a two-layer contact/trust network are converted into biographical profiles using the prosopographic factoid model, assembled into collective narrative via a partner network, and grounded in a documented real-world wind farm dispute. The pipeline surfaces meso-level mechanisms -- broker hesitation, trust-driven convergence, early interaction asymmetry -- that neither aggregate statistics nor raw event logs make visible. The individual-level data that agent-based models already generate serves as the raw material for narrative communication through the output methodology described here.

The purpose of this document is to build on a conversation about how we might communicate confidence in our agent-based models in contexts where they are used for scenario analysis or decision-making. Though confidence is traditionally measured using statistical methods, agent-based models are both software and models. When discussing the results of agent-based models with those who might use insights from them to make decisions or analyse scenarios, assumptions may be made by those using the results about how rigorously the model and its software implementation has been assessed. Rather than stipulating standards to which models should confirm before being used for a particular purpose, we propose to manage end-users' expectations by communicating clearly various levels of rigour in different dimensions by which a model might be assessed. This empowers end-users to make their own minds up about whether and if so how much to trust the results being presented to them.