Cover photo

Detecting real wallet activity in the age of AI bots

Most existing "proof of humanity" solutions ask you to submit a video, scan your iris, or link a government ID. That works for some use cases, but it creates a privacy ceiling and a centralization risk.

Proof of Human

The problem

Airdrop farms. Governance manipulation. Sybil attacks. In 2026 a single operator can deploy thousands of wallets that look indistinguishable from real users — same transaction patterns, same token balances, sometimes even the same NFTs. The difference between a human wallet and a well-scripted bot is increasingly hard to see with the naked eye.

Most existing "proof of humanity" solutions ask you to submit a video, scan your iris, or link a government ID. That works for some use cases, but it creates a privacy ceiling and a centralization risk. It also doesn't scale to the kinds of automated, programmatic checks that protocols and platforms actually need.

POH takes a different approach: instead of asking who you are, it asks what your wallet has done.


How POH works

POH is an open, community-curated network of detection methods — on-chain queries that return a true/false signal about whether a wallet shows evidence of genuine human activity.

A method might check:

  • Whether a wallet holds more than 0.001 ETH on mainnet (low-cost bot wallets often don't)

  • Whether a wallet holds a Binance BAB soulbound token on BSC

  • Whether a wallet's Solana SOL balance has been active for more than 90 days

  • and others.

Any developer can register a new method: an EVM contract call, a Solana program query, or a REST API endpoint. A sandboxed expression evaluator runs user-supplied logic against the live result to produce a boolean outcome. Methods that require the address to be substituted into a URL or body use a {address} placeholder — no server-side template injection is possible.


The AI brain

Running 40+ methods in parallel produces a lot of signal — some conflicting, some weak, some strong. A human would struggle to weigh them consistently.

POH uses a local multi-role Ollama pipeline to synthesise results:

  • Evaluator (DeepSeek R1): takes the top 5 signals ranked by community score, applies per-method weights, runs a second pass to catch overconfidence, and outputs a HUMAN / AI / UNCERTAIN verdict with a confidence score and short technical reasoning.

  • Learner (Qwen 2.5): after every community vote, adjusts the weight of the voted method by ±0.05 (clamped 0.1–3.0). Prevents overfitting to a single vote. Penalises conflicting signals.

  • Compiler (Mixtral/llama3.2): runs hourly, consolidates scan history and vote records into a compact brain state document. No hallucination — only statistically supported facts.

All models run locally. No cloud API. No data leaves the node.


Community incentives

POH is not a protocol run by a single team. Detection quality depends on the community continuously improving and validating signal sources.

Anyone can submit a method for 1000 POH. Once submitted, every scan that uses that method earns the submitter a share of the scan fee.

Scan pricing:

  • First 100 scans per wallet are free

  • Paid scans: 1 POH per address (bulk discounts: down to 0.40 POH/address for 500+ batches)

  • 50% of every paid scan is distributed to method owners, weighted by their stake size.

Community scoring via voting:

  • Any wallet holder can vote on whether a method is a genuine human-vs-bot signal

  • Vote weight is proportional to staked POH

  • The AI Learner updates method weights after each vote — stake more, influence more

Staking:

  • Stake POH tokens to increase voting power

  • No lockup in v1 — stake and unstake anytime

  • Staking contract deployed at BN3bdfPeLTokUVACy9wShMVSCcaF1BqHx5Mg71Ere6CN (devnet)


What POH is not

POH does not ask for:

  • Video submissions or liveness checks

  • Biometric data of any kind

  • Government ID or KYC documents

  • Social media login or email addresses

All verification is wallet-native and pseudonymous. The system never stores PII. The verdict is probabilistic, not authoritative — POH answers "does this wallet show the kind of activity pattern associated with a real person?" not "is this person legally verified?"


Current detection coverage

As of v1, POH runs 44 detection methods across:

  • EVM chains: Ethereum, BSC, Polygon, Arbitrum, Base, Optimism, and 26 more via Alchemy

  • Solana: native SOL balance, SPL token holdings, transaction count

  • REST APIs: ENS reverse records, Farcaster identity, Snapshot governance activity, Gitcoin Passport attestations, Coinbase Verified ID (EAS), Nomis on-chain score, Galxe identity + OATs, Human Protocol oracle, GitPOAP, opBNB BAB

New methods are reviewed by the AI brain for signal quality before being scored by the community.


Token ($POH)

$POH is the utility token that powers fee payment, reward distribution, and stake-weighted governance.

  • Fair launch via bonding curve — no VC rounds, no pre-mine

  • 80% community reward pool (method rewards, staking APY, bonding curve)

  • 20% team/contributors — 1 month cliff, 6 month linear vesting

  • Scan fees flow 50% to method owners, remainder covers infrastructure

  • Devnet token: 9WUxSsvGNTjw5AiqnGc2EmGWwGdgWzfHKawvyk9Bz872


In an era of AI abundance, what remains scarce and valuable is authentic human connection, creativity, and decision-making. Proof of Human ensures we can still recognize and reward it online. Ready to prove you're real?