Brazil2026ElectionFilter: How X-Algorithm Implements Brazil's 2026 Election Content Moderation

The Brazil2026ElectionFilter removes posts authored, retweeted, quoted, or replied to by accounts reported to the Brazilian Electoral Court for the 2026 election from recommendation results, unless the viewer follows those accounts, ensuring compliance with Electoral Resolution No. 23.610.

The Brazil2026ElectionFilter serves as a critical compliance mechanism within the X-Algorithm recommendation pipeline. Located in the xai-org/x-algorithm repository, this Rust-based component operates within the home-mixer crate to filter candidate posts during feed generation. It balances Brazilian legal requirements with user autonomy by preserving content from followed accounts while removing potentially problematic electoral content from algorithmic recommendations.

Purpose and Legal Compliance

The filter enforces Electoral Resolution No. 23.610, Art. 28 §1º-A, which mandates the removal of specific accounts from algorithmic promotion during electoral periods. This regulation targets accounts that have been formally reported to the Brazilian Electoral Court regarding the 2026 election cycle.

Unlike blanket censorship, the implementation respects user agency through the follower exception rule. The filter only removes content when the viewer does not already follow the author, preserving the ability of users to see content from accounts they have explicitly chosen to follow. This architectural decision reflects the nuanced requirements of Brazilian electoral law while maintaining user choice.

How Brazil2026ElectionFilter Works

The filter implements the generic Filter<ScoredPostsQuery, PostCandidate> trait, allowing seamless integration into the candidate processing pipeline. The implementation follows a four-step validation process:

Static Prohibited User ID Set

At the core of the filter lies BRAZIL_2026_ELECTION_USER_IDS, a LazyLock<FxHashSet<u64>> containing approximately 2,500 user IDs. This static set, defined in home-mixer/filters/brazil_2026_election_filter.rs, derives from the open data set of 2026 election candidates. The FxHashSet provides O(1) lookup performance critical for high-throughput recommendation processing.

Author Exclusion Logic

The is_excluded_author(user_id, followed_user_ids) function determines whether a specific author should be filtered. This function returns true only when two conditions coincide:

• The user_id exists within BRAZIL_2026_ELECTION_USER_IDS
• The user ID is not present in the viewer's followed_user_ids vector

This logic enforces the legal requirement while implementing the follow-relationship exemption.

Multi-Relationship Candidate Evaluation

The should_remove(candidate, followed_user_ids) method performs comprehensive relationship checking across four potential vectors of content propagation:

1. Direct authorship – The candidate's author_id is excluded
2. Retweet origin – The original author of a retweeted post is excluded
3. Quote attribution – The author of quoted content is excluded
4. Reply ancestry – Any ancestor user in a reply thread is excluded

If any of these conditions evaluates to true, the candidate is marked for removal from the recommendation set.

Pipeline Integration and Partitioning

Within the filter method implementation, the system extracts the viewer's followed_user_ids from the ScoredPostsQuery context. The candidate vector undergoes partitioning into kept and removed collections based on the should_remove evaluation. The method returns a FilterResult structure containing both sets, allowing the pipeline to track moderation decisions for potential logging or analysis.

Code Implementation and Architecture

Direct Filter Usage

When working with the filter outside the standard pipeline, implementers interact with the Filter trait directly:

use home_mixer::filters::brazil_2026_election_filter::Brazil2026ElectionFilter;
use xai_candidate_pipeline::filter::{Filter, FilterResult};
use home_mixer::models::{candidate::PostCandidate, query::ScoredPostsQuery};

// Construct query with viewer's follow relationships
let query = ScoredPostsQuery {
    user_features: UserFeatures {
        followed_user_ids: vec![12345, 67890],
        ..Default::default()
    },
    ..Default::default()
};

// Create candidate batch containing mixed authors
let candidates = vec![
    PostCandidate { tweet_id: 1, author_id: 1, ..Default::default() },          // Permitted
    PostCandidate { tweet_id: 2, author_id: 14160928, ..Default::default() }, // Blocked ID
];

let filter = Brazil2026ElectionFilter;
let result: FilterResult<PostCandidate> = filter.filter(&query, candidates);

// Verify filtering behavior
assert_eq!(result.kept.len(), 1);
assert_eq!(result.removed[0].author_id, 14160928);

Pipeline Registration

The filter integrates into the Phoenix candidate pipeline through explicit registration in phoenix_candidate_pipeline.rs:

use crate::filters::brazil_2026_election_filter::Brazil2026ElectionFilter;

let filters: Vec<Box<dyn Filter<ScoredPostsQuery, PostCandidate>>> = vec![
    // Preceding filters...
    Box::new(Brazil2026ElectionFilter),
    // Subsequent filters...
];

During request processing, the pipeline invokes each filter's filter method sequentially. The Brazil 2026 filter executes early in the chain, ensuring prohibited content never reaches downstream ranking or feature extraction stages.

Key Source Files and Data Structures

Understanding the filter requires familiarity with these specific files in the xai-org/x-algorithm repository:

• home-mixer/filters/brazil_2026_election_filter.rs – Contains the Brazil2026ElectionFilter struct, static ID set, trait implementation, and comprehensive unit tests
• home-mixer/candidate_pipeline/phoenix_candidate_pipeline.rs – Registers the filter within the candidate processing pipeline at lines 38-41
• home-mixer/models/candidate.rs – Defines PostCandidate, the core data structure filtered, including fields for author IDs, retweet sources, and quote references
• home-mixer/models/query.rs – Defines ScoredPostsQuery, which carries the followed_user_ids vector essential for the exemption logic

Summary

• The Brazil2026ElectionFilter enforces Brazilian Electoral Resolution No. 23.610 by removing specific accounts from algorithmic recommendations during the 2026 election period.
• The filter references a static set of approximately 2,500 prohibited user IDs stored in BRAZIL_2026_ELECTION_USER_IDS within home-mixer/filters/brazil_2026_election_filter.rs.
• Content removal only occurs when the viewer does not follow the author, implementing the legal "follower exception" requirement.
• The should_remove method checks four relationship types: direct authorship, retweets, quotes, and reply ancestors to ensure comprehensive coverage.
• Unit tests verify that the filter maintains non-empty ID sets, correctly handles follow relationships, and properly processes retweet and quote metadata.

Frequently Asked Questions

What specific Brazilian law does the Brazil2026ElectionFilter enforce?

The filter implements Electoral Resolution No. 23.610, Art. 28 §1º-A, which mandates that platforms remove from algorithmic recommendations any content produced by accounts reported to the Superior Electoral Court (TSE) regarding the 2026 elections. The X-Algorithm codebase explicitly references this legal foundation in its documentation and implementation comments.

How does the filter identify which accounts to remove?

The system maintains a static LazyLock<FxHashSet<u64>> named BRAZIL_2026_ELECTION_USER_IDS containing approximately 2,500 blacklisted user IDs. The is_excluded_author function performs O(1) hash set lookups to determine if a user ID requires filtering. This approach ensures minimal latency during the recommendation pipeline's execution while maintaining deterministic behavior based on the official electoral court data set.

Why can users still see content from filtered accounts sometimes?

The implementation respects the follow relationship exception specified in Brazilian electoral law. The should_remove method explicitly checks the viewer's followed_user_ids vector before filtering. If a user follows an account listed in the prohibited set, that account's content remains visible in their recommendations. This architectural decision preserves user autonomy while complying with legal requirements regarding algorithmic amplification of unknown accounts.

Where does the static list of prohibited IDs originate?

The BRAZIL_2026_ELECTION_USER_IDS set derives from the open data set of 2026 election candidates published by Brazilian electoral authorities. According to the source code in home-mixer/filters/brazil_2026_election_filter.rs, this list updates through the standard data pipeline processes and is loaded into memory via LazyLock to prevent repeated disk I/O during high-traffic recommendation requests.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}