social propagation
Mean-Field Propagation Model
This simulator employs a mean-field (aggregated) approach to model viral content propagation across large-scale social networks. Rather than tracking individual agents, the model maintains aggregate state variables for active shares, cumulative reach, and manipulation exposure, evolving them through discrete time steps with branching dynamics modulated by policy interventions. The baseline scenario represents unrestricted propagation, while the policy scenario applies friction mechanisms designed to suppress coordination attacks and reduce the political manipulation surface.
Policy Mechanics
Seven distinct policy levers modulate propagation dynamics: (1) 10-hour cooldown restricts accounts to ~2.4 posts per day; (2) 10-hour coolup delays content visibility, dampening ignition; (3) 1-hour daily posting windows further constrain election-period activity; (4) forward caps reduce effective fan-out by halving the degree; (5) question-gating and prebunking mechanisms lower both share probability and conversion rates; (6) identity tiers impose stricter friction on new/low-trust accounts; (7) per-thread slow mode applies temporal smoothing to growth rates. These interventions collectively reduce the effective reproduction number R_eff, delay cascades, and diminish manipulation impact.
Political Manipulation Impact (PMI)
The PMI metric quantifies expected successful manipulations by accumulating the product of new exposures, per-exposure conversion probability, and attention/skepticism factors at each time step. Conversion probability decays with content age using an attention half-life, penalizing delayed exposure. Question-gating applies an additional skepticism reduction. The model reports absolute PMI for both scenarios and computes percentage reduction under policy, providing a direct measure of manipulation suppression effectiveness. This metric captures not just reach reduction but also the qualitative degradation of manipulation potency due to delays and friction.
Methodological Notes
The mean-field approximation trades agent-level fidelity for computational tractability, enabling multi-day simulations with minute-scale resolution. The model uses logistic saturation to cap reach at audience size and applies a small moderation leak to represent content removal. Slow mode employs exponential moving average smoothing rather than direct R modification. Parameters are calibrated for pedagogical visibility of policy effects rather than quantitative prediction. Real-world dynamics involve heterogeneous agents, network topology, adaptive adversaries, and enforcement imperfections not captured here. Results illustrate comparative trends and mechanism interactions rather than precise forecasts.