Can people coordinate to force a note onto a post (or keep one off)?
The algorithm is specifically designed to resist it, and the math is the reason. Because a note only reaches Helpful status by earning a high intercept, and the intercept only rises when raters across different viewpoint-factor positions agree, a coordinated group all sitting at the same factor position can't push a note through — the model explains their agreement with the factor term, not the intercept. Brigading from one camp is the exact failure mode bridging-based ranking neutralizes. It's not impossible to manipulate, but same-side coordination is structurally weak, by design.
The obvious worry about a crowd-sourced system: can a motivated group game it? Community Notes was built by people who expected exactly this, and the bridging design is the countermeasure. Understanding why same-side coordination fails is the most useful thing in this pillar.
Why a one-sided brigade fails
Suppose a group all sharing one viewpoint floods a note with helpful ratings. In a naive
majority-vote system, that works. In matrix factorization, it doesn't — because the model learns that
these raters all sit at the same factor position, and it explains their agreement with the
factor term, not the note's intercept. The intercept is the part that's common across
viewpoints; ratings from a single viewpoint barely move it.
The model places notes and raters on a shared viewpoint 'factor' axis; a note's intercept only rises when raters across different factor positions agree, and the not-helpful threshold applies a negative note-factor multiplier (default -0.8) so more polarized notes must clear a harder bar. This is the bridging mechanism that resists single-viewpoint agreement.
To push a note's intercept above the 0.40 Helpful threshold, you need agreement from raters at
different factor positions — people the model has learned tend to disagree. A coordinated
single-camp group is, almost by definition, clustered at one factor position, so the one thing they
can't manufacture is cross-factor agreement.
A Community Note becomes Currently Rated Helpful (and shows publicly) when its learned note intercept clears the CRH threshold, which defaults to 0.40 in the open scoring code.
The not-helpful side is hardened too
The reverse — coordinating to keep a deserved note off — runs into the not-helpful
threshold's negative factor multiplier (crnhThresholdNoteFactorMultiplier = -0.8): notes
with larger factor magnitudes must have proportionally lower intercepts to be rated not-helpful. The
asymmetry makes it harder for a polarized group to bury a note than a naive system would allow.
The model places notes and raters on a shared viewpoint 'factor' axis; a note's intercept only rises when raters across different factor positions agree, and the not-helpful threshold applies a negative note-factor multiplier (default -0.8) so more polarized notes must clear a harder bar. This is the bridging mechanism that resists single-viewpoint agreement.
What coordination actually looks like
This doesn't mean manipulation is impossible — it means it has a signature. Genuine manipulation would require either compromising raters across multiple factor positions, or rating patterns that don't match organic behavior: bursts, correlated timing, accounts that rate in lockstep. Those patterns are detectable in the public rating data precisely because they deviate from the viewpoint-distributed norm the model expects.
What the code doesn't say
The production anti-manipulation systems. The open scoring code shows the bridging math that
makes same-side coordination weak, but X runs additional integrity safeguards that aren't fully
published. We can show why the core model resists one-sided brigading; we can't enumerate every
defense layered on top.
The Community Notes scoring algorithm is open-sourced in twitter/communitynotes — a distinct repository from the xai-org/x-algorithm ranking code — and operates on public note and rating data; individual rater data is pseudonymized in the public release.
What to do with this
Trust the bridging property but verify with data: a note that reached Helpful did so by clearing a
cross-viewpoint bar, which is hard to fake — but coordinated rating leaves statistical traces worth
checking. xDoctor's Community Notes intelligence examines exactly these population-level patterns as
an integrity face — surfacing coordination signatures in aggregate, never deanonymizing or targeting
individual raters.
The Community Notes scoring algorithm is open-sourced in twitter/communitynotes — a distinct repository from the xai-org/x-algorithm ranking code — and operates on public note and rating data; individual rater data is pseudonymized in the public release.