A Human Behavioral Baseline for Collective Governance in Software Projects

TL;DR

This work builds a large-scale longitudinal baseline of open-source governance prior to AI-managed tooling by parsing version-controlled GOVERNANCE.md documents into institutional statements using Institutional Grammar. It quantifies how authority and participation evolve through changes in the number and distribution of Roles and Actions (breadth and evenness) while Deontic polarity remains largely stable, using metrics such as $H$, $K$, and $JSD$ across paired snapshots. The findings suggest governance maturation via accretion rather than sweeping reallocation, providing a reproducible framework to evaluate future AI-mediated workflows for concentration, drift, and breadth. The results offer practical guidance for designing participatory AI systems that surface governance deltas and maintain transparent authority dynamics in collaborative software projects.

Abstract

We study how open source communities describe participation and control through version controlled governance documents. Using a corpus of 710 projects with paired snapshots, we parse text into actors, rules, actions, and objects, then group them and measure change with entropy for evenness, richness for diversity, and Jensen Shannon divergence for drift. Projects define more roles and more actions over time, and these are distributed more evenly, while the composition of rules remains stable. These findings indicate that governance grows by expanding and balancing categories of participation without major shifts in prescriptive force. The analysis provides a reproducible baseline for evaluating whether future AI mediated workflows concentrate or redistribute authority.

Figures (3)

• Figure 1: Processing pipeline from raw governance files to structured institutional statements and analysis. These steps support measurement of change in count and concentration from initial to latest versions of version-controlled GOVERNANCE.md project constitutions. This diagram shows how governance text is normalized, parsed into roles, actions, and deontics, and clustered into institutional constructs.
• Figure 2: Latest-snapshot composition of governance constructs. Donuts show the relative share of clusters within each feature (Roles, Actions, Deontics, Objects). These panels are descriptive context; inference relies on paired change metrics reported in the main text.
• Figure 3: Distribution of per-repository distinct construct counts at the initial and latest snapshots for each institutional feature. Violins show density; the horizontal line is the median. Panels provide descriptive context; paired bootstrap estimates are reported in Table \ref{['tab:count-results']}.