Human Passport (passport.human.tech) is the most widely deployed proof-of-personhood layer in DAO funding. It aggregates a scattered pile of credentials – an exchange KYC attestation, a government-ID proof, a biometric check, a Discord account, an ENS name, a record of on-chain activity – into one number, the Unique Humanity Score, which any application can gate on. It was built inside Gitcoin as Gitcoin Passport to stop Sybil attackers farming its quadratic-funding matching pool, spun out as Passport XYZ, and acquired by the Holonym Foundation in December 2024 and rebranded under the human.tech umbrella.
It is the tooling answer to the problem set out in Sybil resistance in DAOs: not who is this person, but how confident are we that this wallet is a distinct human, and what is that confidence worth. Where Collab.Land and Guild.xyz answer “does this wallet hold the right token,” Passport answers the harder question of whether ten wallets holding it are ten people.
A Stamp is a verifiable credential issued by Passport's IAM service after the user proves control of some account or attestation. The catalogue spans five families: KYC, biometrics, web3 activity, web-of-trust, and web2 activity. Each stamp carries a weight, and the scorer “adds up the weights of each Stamp owned by each user and compares the result to a pre-defined threshold value.”
The default threshold is 20. The documentation is explicit that this is a tuned default rather than a law of nature, and publishes the trade-off it encodes:
| Score | Sybil elimination | False-positive risk |
| 20 | Effective | Low |
| 25 | More effective | Medium |
| 30 | Most effective | High |
Integrators are not stuck with it: a custom scorer can move the threshold, upweight or downweight individual stamps, or require a specific stamp as a prerequisite regardless of score. A DAO gating a treasury vote and a grants round gating a $5 donation are entitled to different answers, and the config supports that.
The weights are not a black box – they are a public Python dict in the deployed scorer, gitcoin_passport_weights.py, alongside the threshold constant itself. As currently deployed it lists 51 stamps summing to about 180 points, and the distribution is extremely top-heavy:
| Heaviest stamps | NFTScore#50 16.246 · CoinbaseDualVerification 16.042 · HolonymGovIdProvider 16.026 · BinanceBABT 16.021 · ETHScore#50 16.021 |
| Lightest stamps | Brightid 0.202 · ETHDaysActive#50 0.207 · Ens 0.208 · ETHnumTransactions#100 0.210 · GnosisSafe 0.222 |
Set that against a threshold of 20 and the practical shape of the system is clear. A single Binance or Coinbase attestation carries roughly 80% of the bar on its own. The entire decentralised-identity tail that aggregate personhood was meant to lean on – BrightID, ENS, a Gnosis Safe, a history of Snapshot proposals – sums to about one point between them. Having actually participated in on-chain governance is worth around 1% of the score that certifies you as a human.
This is a defensible engineering choice rather than a scandal: exchange KYC is the strongest and most expensive-to-forge signal available at scale, and the government-ID and biometric routes exist precisely so that people who will not use an exchange have a path. But a DAO adopting the default threshold should be clear about what it is adopting. Aggregate proof-of-personhood, in production, mostly resolves to did a centralised exchange verify you – and that is a different trust assumption from the one the phrase “decentralised identity” suggests. The file is live config and changes; check the current values before quoting them.
Two rules do more work than the score itself.
Deduplication is LIFO. Every stamp carries a hash, and when two Passports present the same credential “the duplicate Stamp is ignored and not counted towards the score” – the first claim keeps it. This, not the threshold, is the actual Sybil defence. Nothing stops one human from creating ten Passports; what stops them is that ten Passports cannot all point at one Coinbase account. Because the heaviest stamps are exactly the scarce ones, every additional wallet has to clear 20 points out of the cheap tail, and the cheap tail does not add up.
Credentials decay. Offchain and onchain stamps alike expire after 90 days, extendable by re-verifying in the app; onchain scores carry no explicit expiry, so the docs advise treating the attestation timestamp under the same 90-day rule. For a one-shot funding round that is irrelevant. For a standing membership roster it is a maintenance obligation – a member verified today is unverified next quarter, and a gate that silently drops people is a governance problem, not just an ops one.
passport-gated validation is still in Snapshot's scoring service (score-api).The stack is open source under AGPL-3.0 – the monorepo carries the app, the iam credential-issuance server, a Ceramic-backed database client, and the infra config, with onchain attestations published across roughly a dozen networks.
Passport was built inside Gitcoin, spun out as Passport XYZ, and acquired by the Holonym Foundation in December 2024, which rebranded it Human Passport and folded it in beside Human Network and Human Wallet. The project's own figures – self-reported, not independently audited – put it at over 2 million users, 150+ campaigns, 120+ integrating projects as of March 2026, nine consecutive Gitcoin Grants rounds defended, and more than $512M in capital flow protected.
The handover is worth noting on its own terms. The Sybil defence that decides who is eligible for a DAO's matching pool is now maintained by a different organisation than the DAO running the pool, on a weights file that organisation controls. That is an ordinary vendor relationship, and it is also a governance dependency: a round's eligibility rules can change without the round's voters approving anything. Anyone treating a Passport threshold as a constitutional parameter should pin the scorer they mean, the way any other external dependency gets pinned.
Caper runs no identity check at all, and makes no claim that any member is human. It attacks the same problem from the cost side instead. A member’s vote weight is earned: it grows with the soulbound vote tokens their own ballots mint, which cannot be bought or transferred, not with the size of their holding alone.
Two things about v matter here, and the second is the one that is easy to get wrong. It is untransferable: its depositor role is bound to the caper’s own state component and the updater is deny_all, so no wallet can send it to another (contracts/core/src/caper_dao.rs). It is not, however, unacquirable. The deployed contract mints v on exactly one surface: casting a ranked ballot mints one (VOTE_MINT, contracts/logic/src/lib.rs), and buying, selling and transferring mint none – the trade-side accrual was removed in the genesis redeploy of 11 September 2026. So v still has a published price, the flat 100 XRD a ballot costs, but the only till that sells it is the ballot box, and it stays in the wallet that voted.
The multiplicative form is what does the Sybil work, and it does it against a cost floor rather than against unbuyability. Splitting one holding across ten wallets multiplies nothing: each wallet then holds a tenth of the tokens, and to reproduce the original weight every one of them would have to accrue its own record from zero – ten times the outlay for, at best, the same weight, while each extra mint inflates V, the denominator every member shares. Extra identities are strictly a worse deal than one, and that holds whether the attacker builds the record by voting or by trading.
It is a narrower promise than proof of personhood: Caper cannot tell you a member is a distinct human, only that concentrated influence had to be paid for in XRD and cannot be consolidated after the fact. But it is a promise the contract keeps by itself, with no external issuer, no expiring credential, and no weights file that a third party can re-tune. See proposals for the voting mechanics.
A note on the two Gitcoin citations below. Gitcoin's entire blog went dark between 13 and 18 August 2026. gitcoin.co still answers 200 at its apex and on every product path, but every URL under /blog now returns HTTP 502 with the body DNS_HOSTNAME_NOT_FOUND, so the edge is proxying to a backend hostname that no longer resolves. The Internet Archive's last successful capture of a post is 13 August 2026 and it has recorded nothing since. The rounds' own eligibility rules survive only in archived captures, which is what the two links point at.
hash field.gitcoin_passport_weights.py – the deployed stamp weights and GITCOIN_PASSPORT_THRESHOLD.