Reputation is compressed history. It lets you get into a stranger’s car, buy from a seller you have never met, or accept code from a maintainer living on another continent. Ten thousand complicated interactions become 4.9 stars, a badge, a commit history or a familiar name.
That compression solves a real problem. Nobody can personally know everyone with whom they need to coordinate.
It also throws most of the truth away.
The score remembers the outcome and forgets the circumstances. It preserves what the platform could measure and discards what it could not. It combines interactions that may not belong together. Then, because the result is legible, it becomes tempting to move it.
Portable to where?
People often say users should own and carry their reputation between platforms. The complaint is sound: a marketplace should not be able to confiscate ten years of honest work by closing an account.
But portability has a destination. A record that says somebody is a reliable open-source reviewer may help another software community. It says little about whether they are a safe babysitter, a careful driver, a good tenant or a wise political representative.
Trust is domain-specific. Reputation systems continually erase that boundary because one number is easier to display than a bundle of conditional statements.
The result is the halo effect when things go well and totalising punishment when they do not. Success in one domain masquerades as character. Failure in another becomes searchable everywhere.
A reputation needs a label
A less dishonest portable record would carry its limits:
- who made the assessment;
- after what kind of interaction;
- in which community and role;
- using which criteria;
- how recent the evidence is;
- what did not transfer.
“Trusted” is nearly meaningless. “Five maintainers accepted twelve of Ana’s security patches in this repository during the past year” is narrower and more useful.
The complexity is not clutter added to an otherwise pure score. It is the information the score removed.
The right to become locally unknown
Good reputation design must also preserve a limited right to start again. Not a right to conceal relevant harm, but a right not to have every old context automatically govern every new one.
Memory needs decay. Translation needs friction. A person should be able to challenge the binding between a claim and their identity. Systems should distinguish direct evidence, hearsay and algorithmic inference.
The aim is not a universal reputation passport. It is a collection of accountable statements whose relevance must be argued at the boundary.
Portability without context gives a person ownership of the cage. The better promise is smaller: your record can travel, but its authority does not travel for free.