Three ideas, and that is the whole thing.
Anyone can write you a perfect email now, so the words cannot be the filter. Kith decides using things an agent cannot mass-produce: who actually knows you, what you personally did with mail before, and what a stranger is willing to spend to reach you.
Nobody has the whole picture.
There is no central list of who is trustworthy. Every person keeps one short, signed list of who they write to, and that is the only thing they are allowed to claim. You collect the lists that reach you and build a private picture nobody else can see. Below is one at real scale: 221 people and 643 relationships, ranked from your seat.
You collect the lists that reach you and build a private picture. Darker means closer to you. You have 26 direct contacts out of 221 people, and everyone else is ranked only by how they connect back to those.
This is also where PageRank earns its keep. The obvious shortcut is to count how many of your contacts know a sender, but that agrees with the full ranking on only 68% of your top 25. Dilan D. shares only 1 contact with you, so counting buries them at number 75, while where they sit in the graph puts them at number 16. A count sees how many. Position sees who.
The score is yours, not theirs.
Karma is how close someone sits to you in that private picture, from 0 to 100. It is computed from your seat, so the same person carries a different number in every inbox. Here is one sender from the network above, scored by two different people who both actually know them, about 5 times apart.
- reaches the inbox
- YES
- of your contacts vouch
- 7
- reaches the inbox
- NO
- of their contacts vouch
- 0
One person, one set of real relationships, two inboxes that disagree by a factor of five. Now throw their mail away and see how far that travels.
Punishment works the same way. Trash a sender and you change only your own view of them, permanently and invisibly to everyone else. This is the part a public reputation score gets wrong: one shared number per address can be brigaded by strangers who have never met you, and it quietly leaks who you talk to. A personal score can do neither.
A stranger can still buy your attention.
Plenty of mail worth reading comes from people who know nobody you know. Locking them out forever is a real cost, not an edge case. So you can post a task instead: do this, and I will look. The catch is that the task has to cost the sender something an agent cannot simply spend a million times.
- They spend
- Money, refunded if you reply
- You get
- A sender who loses something if they wasted your time
Cheap for one good message, ruinous for a hundred thousand bad ones.
Passing a challenge does not add anyone to your graph and does not raise anyone's karma. It buys one message a look, nothing more.
A comment or a signup filters nobody, because that is exactly what automation is good at. Money at risk, real research, or work you wanted done anyway all scale badly on purpose. That is the point. You are not asking for proof of humanity, you are asking for proof that this message was aimed at you.
All of it is open, and none of it reads your mail.
Kith looks at who wrote to whom and when. It never opens a subject line, a body, or an attachment. The scoring engine, the tests, and every number on this page are public.