Most NFT marketplaces rely on curated verification systems. We wanted to see how far you can get by ignoring marketplace metadata entirely and working only with on-chain data.

So we built a Python provenance engine that reconstructs NFT authenticity from transfer history and contract state.

The pipeline is simple:

Pull full ERC-721 transfer history

Rebuild the ownership chain chronologically

Detect structural transfer patterns

Produce a weighted authenticity score

Example transfer reconstruction:

def get_nft_transfers(contract_address, token_id, chain='ethereum'): results = etherscan_request( CHAINS[chain]['id'], "account", "tokennfttx", contractaddress=contract_address, sort="asc" )

    return [
        {
            "from": tx["from"],
            "to": tx["to"],
            "block": int(tx["blockNumber"]),
            "timestamp": int(tx["timeStamp"])
        }
        for tx in results
        if tx.get("tokenID") == str(token_id)
    ]

From there we analyze:

Mint origin legitimacy

Contract source verification

Circular wash loops (A → B → A)

Transfer clustering in short block intervals

Ownership concentration

Counterfeit collections tend to show shallow ownership graphs and tight flip clustering. Mature collections show distributed, temporally organic patterns.

Everything is derived deterministically from on-chain data. No marketplace APIs, no whitelists.

Curious how others would approach wash-trading detection — graph clustering? probabilistic models? Markov processes?