Personalized vs Public Email Reputation: Why Global Scores Get Gamed
Short answer: A public email reputation gives every sender one score shared by everyone. That is easy to game (fake accounts farm it, pile-ons bury rivals) and often wrong (a sender who is noise to the world can be exactly who you want). A personalized score is computed only from your own mailbox, is different for every user, and is provably resistant to fake-account rings.
Two ways to build a sender score
There are only two architectures for scoring email senders by reputation:
- Public / global. Aggregate signals from many users into one number per sender. Everyone sees the same score. This is how domain reputation and "KarmaRank-style" products work.
- Personalized. Compute a separate score for each user from their own graph, seeded on them. Nothing is shared.
They look similar from the outside. Under the hood they behave completely differently.
Why public scores get gamed
A global score is a target. If one number decides whether a sender reaches everyone, then:
- Sybil attack. Spin up many fake accounts, have them interact to inflate the score, and ride it into real inboxes.
- Brigading (the inverse attack). Coordinate a group to mass-flag a rival and bury them for the entire network.
This is not a patchable bug. Cheng and Friedman proved that a symmetric, global reputation function cannot be made fully Sybil-proof. The structure invites manipulation.
Why personalized scores resist it
When the score is seeded on you, the math changes. A random walk that always restarts at you only assigns trust to senders with a real path from you. A ring of fake accounts can link to each other as densely as it likes. With no genuine edge into your graph, it stays near zero:
Clique S with no path from you ⇒ π_v ≈ 0 for every v in S
In simulation, this holds against thousands of fake accounts. The attacker would have to earn a real relationship with you, which is exactly the costly, human thing the system is designed to reward.
Public scores are also just wrong
Gameability aside, a global number cannot represent the most important case in email: a sender who is noise to the world but gold to you. Your accountant, a founder you backed, a former colleague. Their public score would be low. Their personal score with you is high, and that is the correct answer. One number for everyone erases it.
What about negative signals?
Marking a sender spam should lower their score. A public system makes that flag global, which reopens the brigading attack. A personalized system keeps it personal: your spam flag craters your view of that sender and never touches anyone else's. If cross-user distrust is ever useful, it should flow through trusted paths, not a raw public tally.
FAQ
Is domain reputation a public score? Yes. SPF, DKIM, DMARC, and ISP domain reputation are global. They prove a domain is not forged, but they say nothing about whether you want the mail. See authentication is not trust.
Does personalized mean my data is shared? No. The whole point is that your graph and your score never leave your account.
Where can I read the math? See PageRank for email.