Every professional network has a reputation system, and none of them work. Follower counts measure how interesting someone is to watch, which is nearly unrelated to whether they deliver. Endorsements measure social reciprocity — I click yours, you click mine — and cost nothing, so they mean nothing. Recommendations are written by friends and read by nobody.
The result is a status layer that is easy to game and useless for the one decision it should inform: should I bring this person onto something that matters to me?
Start from what you actually need to know
Before working with someone, the questions are narrow. Have they finished things? Did the people who worked with them say it went well? Do they answer when contacted? Have they created value for others, or only extracted it?
None of that is captured by a follower count. All of it is capturable.
Confirmation is the whole game
The single design decision that separates a real reputation system from a decorative one: can a member's claims count without anyone else agreeing?
If yes, the system measures confidence, not competence. The loudest self-promoter wins.
So outcomes need a second party. A member reports that a collaboration produced revenue; the other participants confirm or dispute it. Unconfirmed outcomes stay visible — they are not deleted or hidden — but they never count toward any headline number. It is a small rule with large consequences: it makes the top of the leaderboard expensive to reach dishonestly, because it requires accomplices.
Designing against the obvious attacks
Any scoring system will be gamed the moment it matters. The attacks are predictable, so the defences should be structural rather than reactive.
Endorsement rings. Four people who only endorse each other look impressive by volume. Weighting endorsements by how reciprocal and how clustered they are collapses that to nearly nothing.
Sockpuppets. Fake accounts confirming fake outcomes. Confirmations should only carry weight from members who have independent standing of their own — new accounts cannot manufacture status for anyone.
Purchase. The simplest defence: build no path. Paid tiers can unlock features; they must never grant standing. If there is no mechanism for reputation to change hands, there is nothing to buy.
Volume spam. Ten thousand tiny contributions should not outrank one delivered contract. Diminishing returns per category, per period.
Staleness. Something delivered in 2019 says less than something delivered last quarter. Recent activity should weigh more.
Keep the components visible
A single composite score is convenient and slightly dishonest — it hides what it is made of, and hidden things get gamed. Showing the parts alongside the total means a member with a high score built on endorsements looks different from one built on completed contracts, and anyone deciding whether to work with them can see which.
The point of reputation is not to rank people. It is to let a stranger make a good decision quickly. Any component that does not serve that is decoration.