Coverage & scoring
We track accounts whose current Bluesky profile shows a valid verification badge.
How accounts are found
- Find trusted verifiers. We check known verifier profiles and discover additional candidates from valid verification records.
- Check their records. We page each trusted verifier’s public repository for accounts it has verified.
- Confirm each badge. We check every account against Bluesky’s current profile status; only valid badges enter the tracked directory.
A live verification stream catches changes, and daily checks reconcile the roster. Bluesky has no public global list of every trusted verifier, so this discovered directory may be incomplete or lag behind changes.
What we score
Only standalone original posts qualify. Replies, reposts, and quote posts are excluded. Each eligible post gets two automated estimates: a harmful signal (abuse, bad-faith outrage, or materially misleading claims) and a constructive signal (empathy, useful information, or good-faith dialogue).
- First 30 eligible posts per account: scored without sampling.
- After that: up to 3 posts per day, at most 1 in each 8-hour UTC window.
- Each post is assessed once. Retries and edits do not trigger a second assessment.
- Signals count at 0.70 or above. Scores are estimates, not findings about a person.
More about checks and data
Account verification is rechecked daily. Pages show the sample size and last verification check. The model receives post text only (up to 300 grapheme clusters); links and media are not sent. After scoring, post text is discarded; only the post URI, timestamps, and numeric scores are retained. Blue Watch is not a moderation or enforcement service.