3 min read Measurements
Who holds the keys of ERC-3643 token issuances?
We listed every token announced by the standard ERC-3643 factory event on Polygon and Ethereum, read its owner, and counted. Anyone can recount it with one command.
We counted. Using only public chain data, we listed every token announced by the standard ERC-3643 factory event on Polygon and Ethereum (3,427 issuances from 222 factory contracts), read the owner of each one, and followed that owner until it reached an account, a Safe, a timelock or a dead end.
The result: 97.2% of the issuances have a single account, or a Safe that needs only one signature, as owner. 0.3% need more than one party (a Safe with two or more signatures, or a timelock that answers getMinDelay()). Among the 1,052 issuances that are active (supply above zero, not paused), the shares are 95.7% and 0.8%. No owner in the list answers getMinDelay(), the call that identifies a standard timelock.
The numbers
Counted on 16 November 2026 from public chain data. Each number can be recounted with one command.
3,427
ERC-3643 issuances announced by the standard factory event, 3,363 on Polygon and 64 on Ethereum, from 222 factory contracts.
data/discovery.json97.2%
have a single account, or a Safe that needs one signature, as owner.
README, the numbers0.3%
need more than one party: a Safe with a threshold of two or more, or a timelock that answers getMinDelay().
README, what was measured1,052
issuances are active. 95.7% of them have a single key and 0.8% need more than one party.
data/summary.json80%
of the active issuances have one of the 50 largest of 222 final controllers as final controller, counted per issuance and not by supply.
data/tokens.csv
Agents and levels
On 17 November 2026 we also read who holds the agent role. In ERC-3643 an agent can mint, burn, freeze and force transfers on its own, and the owner can add and remove agents. 789 distinct agent addresses hold the role somewhere in the list. 86.2% of all issuances have at least one agent that is directly an account or a Safe that needs one signature. Counting also agents that are contracts whose owner chain ends at an account, the share is 91.9%.
We then took the weakest role holder, the owner or any agent. 98.5% of the issuances are at Level 0, where at least one role holder is a single address that can act alone. 0.3% are at Level 1, where every role holder is a Safe that needs two or more signatures. A level describes who can act alone, not whether an issuer is safe, and no issuer is named. The definitions are in LEVELS.md in the repository, and the command node card.mjs followed by a token address shows the owner, the agents and the level of one token in words.
What we do not claim
An account address can sit in front of custody software such as MPC or an HSM, which cannot be seen on a chain, so a single key here means a single address. The agent numbers are one snapshot taken on 17 November 2026 and are not part of the weekly series yet. Agents of the identity registry are not read, and neither are upgrade rights or anything off chain. Anyone can deploy a factory that emits the event, so the list includes small and test deployments, and supply is not money. A Safe with two or more signatures is counted as multi party even if a module lets fewer signers act. This is an observation about who can act. It does not say that anything is wrong, and it names no issuer or platform.
Why we publish it
A number nobody can check is worth little. The code that builds the list, the code that counts, the data and a recount script are public, and the recount script plants at least thirty errors in a copy of the data and fails if it misses any of them. If you count differently, please tell us and we will credit you and correct the page, in the open.
Cite as: holistis (github.com/holistis), ERC-3643 key-control census, 16 November 2026. DOI 10.5281/zenodo.24323265.
Recount it yourself
The repository has the data, the code that builds the list, a one-command recount and a command that shows one token in words. It needs Node 22 and nothing else.