Almost everyone has a reputation online, and almost none of it belongs to them. A seller’s ratings sit with the marketplace. A freelancer’s reviews sit with the platform that arranged the work. A player’s record sits with each game, ladder and server they have ever played on. Each fragment is real, and each one disappears the day the person leaves or the platform closes. This is the last piece in our series on identity and data, and it covers what a portable reputation needs: one the person holds, takes with them, and can prove to someone who has never heard of the platform that produced it.
Platform ratings inflate until they stop meaning anything
The first problem with platform-held reputation is that it decays. The economics research on this is consistent. In a large eBay dataset studied by Chris Nosko and Steven Tadelis, the median seller had a one hundred percent positive feedback score, which means the score had stopped telling buyers anything about who was better than whom (NBER working paper 20830). On one large online labor platform, the share of contracts rated exactly five stars rose from 33 percent to 85 percent over six years, and Apostolos Filippas, John Horton and Joseph Golden found that more than half of that rise was inflation rather than better work (NBER working paper 25857).
The cause is structural. On a platform, a generous rating costs the rater nothing and a harsh one invites conflict, so ratings drift up. The platform has little reason to stop the drift, because high scores make the marketplace look healthy. Over time the measure stops separating anyone from anyone.
A portable reputation, held by the person and built from verifiable records and checked by a party with no stake in the outcome, changes that. When the record has to stand up in front of a verifier who gains nothing from flattering anyone, it is built from facts that are hard to inflate.
Two attacks that are identity problems first
The second problem is that reputation is only as good as the link between the record and the person. Two attacks matter most, and neither one is solved by a better scoring formula.
The first is one person appearing as many. If accounts are cheap, a bad actor can build several reputations, or hide a bad one behind a fresh account. The second is a record that does not match whoever is doing the work today: a strong history on an account that has changed hands.
Both matter more every year, because reputation is increasingly what gets bought. A tournament admitting entrants, a guild handing out a role and an AI team paying people to judge model outputs are all buying a claim about a specific, distinct person. Our work on proving one unique person without disclosing which one came out of exactly this problem. A portable reputation has to be bound to an identifier the person controls, so the record and the person cannot quietly come apart.
What a portable reputation needs from the infrastructure
Last week we walked through how Ontology’s decentralized identity works. For reputation, three properties of that layer do most of the work.
Checks that do not go back to the issuer. A verifier can confirm who issued a reputation credential, that it has not been revoked, and that the person presenting it controls the identifier it was issued to, all by reading Ontology’s public chain. None of those checks involves contacting the platform that produced the record, so the record still verifies after that platform has gone.
An order of events nobody can rewrite. A reputation claim is worth checking partly because of when it was made relative to other events: a ban, a lost key, a change of rules. An issuer that controls its own timestamps can backdate. A shared, append-only record cannot be quietly reordered, which is the difference between a claim about the past and a record of it.
Revocation the verifier can read. If a credential is withdrawn, a status list lets every later verifier see that without the issuer learning who checked.
Behavioral data stays off the chain
The industry has spent years showing how not to do this, mostly by writing people’s behavior to public ledgers and calling it ownership. Ontology’s rule is strict. No score goes on chain, and neither do game histories, ban records, ratings, evaluation outputs or any identifier that leads back to a person’s real-world identity. Hashing or encrypting that data does not make it safe to publish, because a public ledger is permanent and encryption ages.
The chain holds only what a stranger needs to check a claim: identifiers and their keys, the issuer’s registry entry and revocation status. The reputation itself is a verifiable credential held by the person. The evidence behind it stays wherever it was generated, under whatever retention rules apply. That boundary is also what keeps a portable reputation workable where people have a legal right to have their data erased.
Gaming is the first test
Gaming is where Ontology has taken this furthest, because players already have deep records and a reason to carry them. The gaming reputation work is built around the parts of a record that are hard to fake: achievements few players have unlocked, rather than hours logged; a dated conduct record; and standing that other people granted, such as moderator roles and community trust. The player holds the result as a portable reputation credential and chooses, item by item, what a tournament or guild gets to see. Each disclosure is time limited and can be withdrawn.
A portable reputation helps anywhere a person’s track record matters to a stranger: a freelancer moving between platforms, a contributor joining a new project, an evaluator taking on paid work. Gaming is where the audience was ready first.
The test for any reputation system is one question a verifier should be able to answer without contacting anyone: is this credential valid now, and was it valid when it says it was? If the answer needs a call to the company that issued it, the reputation still belongs to that company. On Ontology, it belongs to the person it describes.
Further reading: from May, Portable Reputation: The Missing AI Evaluator Infrastructure. The identity and data series: our Korea Blockchain Week recap, zero-knowledge KYC and the Peirce speech and how Ontology’s decentralized identity works.
