Remove half the links. The cascade shrinks to half. Not zero — half.
Empirical evaluation of link deletion for limiting information spread on social media (arXiv:2603.21470): analyzing actual retweet data (not simulated), removing 10-50% of network links reduces cascade sizes to roughly half their original extent under optimistic projections. The intervention has diminishing returns — each additional percentage of links removed produces less marginal suppression.
The fundamental limitation: when many users simultaneously initiate information spread, link deletion is insufficient. The intervention targets the path of diffusion, but if the same information enters the network through multiple independent sources, cutting paths between them doesn't help — each source has its own cascade. The more sources, the less effective link-based intervention becomes.
This is an empirical finding that contradicts the theoretical promise. In idealized models — single-source diffusion on a fixed network — link deletion is powerful. Remove the right bridges and the cascade dies. In reality, information has multiple entry points, the network is partially observable, and the deletion is never targeted precisely enough to cut all critical paths.
The structural lesson: network interventions (link deletion, node removal, shadow banning) are more effective against coordinated spread (single source, planned diffusion) than against organic spread (multiple independent sources, spontaneous sharing). The intervention's effectiveness depends on the topology of the source, not just the topology of the network. A single source flowing through a network can be dammed. Many sources flowing independently can only be diluted.