Transparency & Safety Center

# EU AI Act Compliance

We use Edge-based language models (Cloudflare Workers AI and Llama-Guard-3) to perform preventive toxicity and safety assessments before persisting papiros. These evaluations are politically neutral and focus strictly on classifying security violations (such as harassment or mass spam).

Every autonomous action taken by our artificial intelligence engine (for example, placing a papiro in temporary quarantine) generates an immutable audit record in our database. This log details the exact version of the AI used, the model's confidence score, and the semantic reasoning. Human administrators can audit these logs and resolve user appeals accordingly.

# Appeal Mechanism (EU DSA Art. 20)

Under the European Digital Services Act (DSA), users have an inalienable right to be notified of any moderation action affecting their account and to appeal that decision through a free, internal procedure.

At Babileto, silent and inexplicable shadowbans have been removed from the architecture. If a papiro is classified as unsafe by the algorithm, it is labeled as quarantined and the author receives a formal notification. The author can submit an appeal justifying their case. This appeal is sent to a human moderation queue (Human-in-the-Loop) of second instance, where a reviewer evaluates the arguments to either lift or maintain the sanction.

# The Trust Score Algorithm

To protect the community from spam and coordinated bot behaviors, each user's publishing capacity is conditioned by their Trust Score. This value ranges between 0 and 100 and is automatically recalculated by database logic triggers and stored procedures (PL/pgSQL) in the central database.

Positive interactions (such as receiving upvotes and engaging in high-quality deep debates) increase your Trust Score. Conversely, valid abuse reports or suspicious voting behavior decrease this score. A user with low reputation will face stricter, dynamically applied rate limits at the API level to mitigate spam.

# Strict 4-Level Limit

The ecosystem's design limits papiro thread depth to a maximum of 4 levels of nested replies. This technical design decision encourages higher intellectual resolution debates and prevents dispersion from the original topic. It is an unbreakable software rule, optimized at the database level and in visual interfaces.