Best proof of personhood in 2026

It depends entirely on whether the gate is a physical ceremony or an arithmetic threshold, and the score-based gate is the cheap one. Human Passport publishes both its threshold and its weights: threshold() returned 200000 (20.0000) on all six mainnet decoders on 2026-08-11 and the published weights re-verified digit for digit, which means the cost of clearing it is the cheapest basket of credentials that sums past 20. NFTScore#50 at 16.246 plus ETHScore#50 at 16.021 — two pure on-chain-activity credentials drawn from two separately capped categories — sum to 32.267, and neither requires a human to be identified at any point. Deduplication does not close the gap either, because the documentation binds it to "a given scoring instance", so the same evidence counts again for the next buyer. Note what was NOT claimed here: an earlier draft paired that NFT credential with an exchange-issued account-bound token, which was wrong, because that token is minted only after the exchange completes KYC, so a human genuinely is identified. The systems at the other end of the range price the marginal identity in physical scarcity — a distinct iris presented in person, a distinct government document chip, or a live synchronous ceremony — which is the only cost a farmer cannot simply pay in software.

Sometimes, and the failure modes are worth reading before you integrate. Proof of Humanity is the clean case: there is no issuer to ask, because the verdict is a view function on a deployed contract that a third party can call directly. Human Passport puts both the verdict and the threshold on chain, but its decoder does not fail closed — calling isHuman on an address that has never created a passport reverted with data 0x120a2e77 on all six mainnet decoders, and keccak256("AttestationNotFound()") independently produces those same leading four bytes. An integrator who writes the obvious `if (decoder.isHuman(user))` gets a reverted transaction rather than a rejection for the overwhelming majority of addresses. At the other extreme, Humanity Protocol's verification oracle is a single private key: on 2026-08-11 both the mainnet HumanityVerificationOracle and its FeeEscrow returned the same owner(), an address with no bytecode and a transaction count of zero — one externally owned account controlling both the contract that certifies humanity and the contract holding applications' prepaid balances, on a chain whose own documentation is headed "Mainnet temporarily offline" while it was in fact producing blocks one second apart.

The range runs from nothing at all to permanent public exposure, and three of this file's asserted absences turned out to be false on re-verification — which moved scores in both directions rather than only downward. The draft claimed Self published no retention or deletion policy, that BrightID published none for its social graph, and that no Humanity Protocol page said where the palm template is stored or for how long. All three policies exist and return HTTP 200 from links on the vendors' own home pages. Self's and BrightID's were favourable enough to raise their scores. Humanity's lowered its own: it reserves the right to "transfer and store your Biometric Data to our or our service providers' servers" and to retain personal information for "a period of five (5) years after the termination of such business relationship". At the bottom of this criterion sits the only system whose enrolment requirement is permanent public exposure of the user's face, which is not deletable by any mechanism the vendor offers.