reducing algorithmic recommendation for content approaching but not clearly violating guidelines — is among the most consequential and least transparent moderation interventions at scale. Its existence illustrates both the potential for soft moderation to reduce harmful content reach and the transpa