Community Notes

Context from people who disagree about everything else.

The idea

Any logged-in user can write a note on any listing. Notes are contextual annotations — “this tool requires a paid API key to work,” “the author abandoned this project in 2024,” “there's a known security issue in version 2.x.”

But a note doesn't show up just because someone wrote it. It shows up when the algorithm determines that people who normally disagree both find the note helpful.

The bridging algorithm

The algorithm uses a technique called matrix factorization. Here's the simplified version:

Step 1 — Build a rating matrix

Every user who rates a note (helpful / not helpful) becomes a row. Every note becomes a column. The matrix captures the pattern of who finds what helpful.

Step 2 — Find the axes of disagreement

The algorithm discovers dimensions along which raters cluster — people who tend to agree with each other form groups. These groups naturally emerge from voting patterns, not from any labels or identities.

Step 3 — Surface what bridges the gap

A note scores high when users from different clusters all rate it as helpful. The helpfulness score is essentially: what's agreed on across diverse raters, after accounting for group bias.

The formula (simplified)

helpfulness = intercept
  + (rater_factor · note_factor)
  + note_intercept

A note is shown when note_intercept > threshold

The note_intercept captures what raters agree on
AFTER removing the variance explained by group
alignment (rater_factor · note_factor).

In plain language: strip out the predictable disagreement, and what's left is genuine consensus. That's what gets shown.

Why this beats majority voting

Majority voting has a fundamental problem: the majority can suppress minority viewpoints. If 60% of users on a platform share a particular perspective, they can vote down any note that challenges it, even if the note is factually accurate.

The bridging algorithm sidesteps this by requiring cross-group agreement. A note written by someone in the majority only gets shown if people in the minority also find it helpful, and vice versa. This makes it extremely resistant to brigading and pile-ons.

Open source

This algorithm was pioneered by Twitter/X's Community Notes team and the full implementation is open source.

twitter/communitynotes on GitHub

We ship with this from day one.

Most platforms bolt on moderation after problems appear. walnut.world launches with the bridging algorithm built into the foundation. Community Notes isn't a feature — it's the architecture.