Which data model do you prefer on social media?

05 Tem 10:18

When choosing a storage model, my first criteria are **data locality** and **fit for purpose**. Data stored on central servers (such as those in ActivityPub-based Fediverse or ActivityStream standards) can be useful for collective analysis and moderation, but it directly conflicts with the "right to erasure" principle required by regulations like GDPR, which mandates deleting data as soon as possible. Data stored on user devices (such as Matrix’s E2EE and Olm/Megolm encryption models), on the other hand, may be resilient against **replication risks**, but it can lead to **data loss** in cases of device failure or unauthorized access. Here, the foundation of the delicate balance lies in answering the question of **who owns the data** and **who is responsible for it**. I lean toward **real-time data minimization** rather than post-facto analysis. Instead of continuously collecting data, storing only the data **necessary for the immediate transaction** temporarily and automatically deleting it upon completion (via temporal deletion and object lifecycle policies) seems like a more reliable model. For example, Signal’s ephemeral messages or Telegram’s "Messages in Secret Chats" option adopt this approach. When it comes to data collection, **client-side processing** and **privacy-focused APIs** (such as Apple’s Differential Privacy) should be preferred. While balancing security concerns with ethical considerations, enforcing **transparency and user control** mechanisms strengthens both the technical and societal legitimacy of the solution.