Code-Geometry ABM: Endogenous Enforcement Delegation in Religious Communities

An agent-based model of how geometric symmetry in costly signaling produces endogenous enforcement delegation and punishment concentration in religious communities -- without pre-assigned enforcer roles.

Agents on a scale-free network cultivate internal belief (ITC) and visible ritual practice (EPC) under periodic external threat. When threat is high, institutional authority rises and a monopoly-on-enforcement mechanism activates: a small cadre of high-literalism agents emerges as enforcers through patronage and institutional capital accumulation, concentrating >80% of punishment acts in <5% of the population. This concentration is not designed in -- it arises from the interaction of costly signaling legibility, exit barriers, and threat-driven authority.

Repository layout

src/                    model code (12 versions, v0 through v2.7)
scripts/                run scripts, sweep drivers, aggregation, validation
results/                committed result sets (CSV data + publication figures)
  v2.5_*/               confirmatory sweep, ablation, baseline (5 dirs)
  v2.6_*/               doctrinal retention module + ablations (5 dirs)
  v2.6b_*/              membership module + ablations (5 dirs)
  figures_publication/   publication-ready PNG + PDF figures
AUDIT_REPORT.md         independent code audit

What is NOT in this repo

• runs/ (raw simulation output, ~5 GB) -- gitignored; regenerate via scripts
• tmp/ (scratch) -- gitignored
• .venv/ (virtual environment) -- gitignored; recreate via requirements.txt
• v2.7* results -- experimental; not part of the current manuscript

Quick reproduce (figures from committed CSV only)

This regenerates all publication figures from the committed CSV data. No simulation reruns required. Takes ~30 seconds.

git clone https://github.com/khatvangi/code-geometry-abm.git
cd code-geometry-abm
./reproduce.sh

Or step by step:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

python scripts/build_v2_5_publication_figures.py \
  --root results/v2.5_corrected_three_regime_confirm \
  --outdir results/v2.5_corrected_three_regime_confirm/figures_publication

python scripts/validate_figures_from_csv.py

Path portability note

Some CSVs store absolute paths from the original run environment (the run_dir column in sweep_seed_results.csv). reproduce.sh rewrites these path prefixes to the local clone path on first run; this affects paths only, not numeric data. A backup is saved as .csv.bak before any rewrite.

Full rerun (expensive -- not required for review)

A full confirmatory sweep (72 parameter cells x 5 seeds = 360 runs) takes ~4-6 hours on a 64-core machine. Each run produces ~450 timesteps of 300 agents on a scale-free network.

# v2.5 confirmatory sweep
python scripts/run_v2_5_corrected_sweep.py

# v2.6 retention sweep + ablations
python scripts/run_v2_6_retention_sweep.py

# v2.6b membership sweep + ablations
python scripts/run_v2_6b_membership_sweep.py

Do NOT rerun unless you need to verify stochastic reproducibility. All committed CSVs were generated with fixed seeds and are deterministic for the pinned dependency versions.

Key outputs

Paper section	Result directory	Key file
Phase map (Fig 1)	v2.5_corrected_three_regime_confirm/	phase_bundle/phase_table.csv
Regime counts (Fig 2)	v2.5_corrected_three_regime_confirm/	phase_bundle/regime_counts.csv
Punishment concentration (Fig 3)	v2.5_corrected_three_regime_confirm/	sweep_seed_results.csv
Regime trajectories (Fig 4)	v2.5_corrected_three_regime_confirm/	timeseries_regime_hier_*.csv
Literalism enrichment (Fig 5)	v2.5_corrected_three_regime_confirm/	per-seed agent_summary.csv
Baseline (30 seeds)	v2.5_corrected_baseline/	regime_by_seed.csv
Fixed-y0 ablation	v2.5_ablation_fixed_y0/	ablation_comparison.csv
Retention module	v2.6_validation/	validation outputs
Membership module	v2.6b_validation/	validation outputs

Publication figures: results/v2.5_corrected_three_regime_confirm/figures_publication/

Model versions

File	Key addition
religion_fundamentalism_abm.py	Base model (v0): costly signaling on networks
_v1_1.py	Minor refinements
_v2.py	Heresy/apostasy dynamics
_v2_1.py	Exit friction parameter
_v2_2.py	Regime grid exploration
_v2_3.py / _v2_3_1.py	Mixed regime search, boundary refinements
_v2_4.py	Internal army / enforcer delegation
_v2_5.py	Monopoly on enforcement, literalism trait (primary manuscript version)
_v2_6.py	Doctrinal retention module
_v2_6b.py	Membership benefit module
_v2_7.py	Endogenous delta (experimental)

Python version and dependencies

• Python 3.12.4 (see .python-version)
• Dependencies pinned in requirements.txt
• Core stack: Mesa 3.4.2, NumPy 2.4.2, pandas 3.0.0, NetworkX 3.6.1, Matplotlib 3.10.8

Citation

If you use this code, please cite:

@software{khatvangi2026codegeometry,
  author    = {Khatvangi, Kiran},
  title     = {Code-Geometry ABM: Endogenous Enforcement Delegation in Religious Communities},
  year      = {2026},
  url       = {https://github.com/khatvangi/code-geometry-abm},
  note      = {DOI forthcoming}
}

License

MIT License. See LICENSE.