Keith Coleman is Product Lead for Community Notes at X/Twitter. Jay Baxter is the Founding ML Engineer and Researcher who designed the bridging-based agreement algorithm at its core. Community Notes is X’s public fact-checking system: any user with a verified phone number can become a contributor, propose notes on posts they find misleading, and rate notes proposed by others. A note appears publicly only when contributors who normally disagree with each other all rate it helpful — a condition structurally resistant to manipulation.
This episode will anchor a Community Notes concept page.
Key ideas
- Bridging-based agreement, not majority rules: notes appear only when contributors who disagree politically converge on rating them helpful. Partisan manipulation cannot produce the cross-partisan agreement the algorithm requires, so notes that appear are structurally neutral. All code is open source.
- The founding assumption was wrong: when Baxter built the algorithm in 2020, every ML engineer assumed fact-checking required ground-truth labels from professional fact-checkers. The bridging algorithm needs none — it infers quality from the pattern of disagreement and agreement itself.
- Survived three CEOs: Community Notes (originally Birdwatch) passed through Jack Dorsey, Parag Agarwal, and Elon Musk intact. The algorithm’s demonstrable neutrality made it politically untouchable at every leadership transition.
- No one is exempt: Musk himself, government officials, and advertisers have all received Community Notes. The team treats exemptions as existential threats to the product’s credibility.
- Meta adopted it: independent studies confirm that posts with Community Notes measurably reduce readers’ stated agreement with misinformation claims. Meta adopted the system as its primary fact-checking tool, replacing professional third-party fact-checkers.
How Community Notes works
Any X user with a verified phone number can apply to become a contributor. Contributors propose notes on posts they judge misleading and rate notes proposed by others on a helpfulness scale.
The bridging-based agreement algorithm — designed by Jay Baxter — decides which notes appear publicly. The core mechanism: the algorithm identifies contributors who normally rate things differently from each other (a proxy for political or ideological disagreement). A note only passes the threshold when contributors from across that disagreement spectrum all rate it helpful.
This produces three properties simultaneously:
- Neutrality: a note that only partisan contributors rate helpful never appears. The cross-partisan requirement filters it out.
- Quality: notes that clear the bar tend to be accurate, well-cited, and informative — contributors who disagree on politics still agree on what a good note looks like.
- Manipulation resistance: coordinated manipulation requires mobilising contributors who disagree with the manipulators. Bad actors can flood the system with one-sided ratings; the algorithm discards them for lacking cross-partisan corroboration.
The algorithm requires no ground-truth labels from professional fact-checkers. It bootstraps quality from the structure of disagreement itself — the non-obvious founding insight. All algorithmic code is open source; anyone can audit how a note was scored.
Origins and surviving leadership changes
Birdwatch began as an internal contest at Twitter in 2020. Baxter built the bridging algorithm as his contest entry. No one inside Twitter believed public fact-checking could work without professional moderators; the internal assumption was that the crowd would produce noise, not signal.
The product launched under Dorsey, survived Agarwal’s tenure, and continued under Musk. At each transition the team faced potential shutdown or restructuring. The algorithm’s neutrality was its political asset: no CEO could credibly argue that a system requiring cross-partisan agreement was biased against any faction.
Kayvon Beykpour, who championed Community Notes internally, used his two-hour session with Musk after the acquisition to argue for keeping the team. Musk continued the product. Under Musk the team ran leaner — Coleman noted they accomplished more with fewer people than before.
Impact
Independent external studies show posts with Community Notes measurably reduce readers’ stated agreement with misinformation claims. The effect replicates across studies.
Meta adopted Community Notes as its primary fact-checking tool, replacing its previous system of professional third-party fact-checkers. For a product that began as an internal contest entry in 2020, adoption by Meta validates both the algorithm and the premise that bridging-based agreement generalises beyond X’s user base.
The product’s operating principles remain fixed: all of humanity participates, not a selected expert class; notes must be informative rather than merely critical; no contributor or subject is exempt.