Token-based governance appears to be a single process but is usually two. Voting happens off chain because it is cheap there, while execution must happen on chain because that is where the effect exists.
On-chain voting costs money per voter
Recording a vote in a contract is a transaction, so every participant pays a fee to have their preference counted.
For holders with small balances the fee can exceed the influence being exercised, which discourages exactly the broad participation governance is supposed to demonstrate.
Turnout falls as a result, concentrating outcomes among large holders who can afford the cost relative to their stake.
Signed messages cost nothing
A signature over a structured message proves that the holder of an address approved specific text without submitting anything to the network.
Off-chain platforms collect these signatures and weight them by token balance measured at a snapshot block, producing a verifiable tally at no cost to voters.
Using a past block prevents tokens being borrowed or moved between wallets to vote more than once with the same balance.
The tally does not do anything by itself
A signature collection is evidence of preference. Changing a parameter, moving treasury funds or upgrading a contract still requires a transaction.
Someone must therefore construct and submit the transaction implementing the result, and that party has to be authorized to do so.
Authorization typically rests with a multisignature wallet whose signers are expected to execute what the vote decided.
The gap between vote and execution is a trust assumption
Signers could execute something different, or decline to execute at all, and the off-chain vote provides no protocol-level constraint preventing it.
The practical safeguards are reputational and organizational: publicly known signers, published transactions and observers who compare what was executed against what was approved.
Hybrid designs narrow the gap by having the off-chain result submitted with a proof, so the on-chain contract verifies the tally rather than trusting an intermediary.
Timelocks make the result reviewable
Many treasuries route approved actions through a timelock contract, which queues a transaction and permits execution only after a delay.
The delay exists so that anyone can inspect the queued transaction and react before it takes effect, including by exiting if the action is not what was described.
Timelock duration is therefore a governance parameter in its own right, trading responsiveness against the time available for scrutiny.
Protocols commonly exempt emergency actions from the delay, which reintroduces the discretion the timelock was meant to constrain and makes the definition of an emergency worth reading closely.