Delegation is the mechanism that makes large token electorates workable: a holder who will never read a proposal assigns their weight to someone who will. Voting and delegation covers how that market forms and re-concentrates, and voting power activation covers the supply-side arithmetic underneath it. This page covers what happens afterwards – the accountability half. A delegate who stops voting does not return the weight, does not stop being counted, and in most DAOs is not removed by any mechanism at all. The holders behind them usually never notice, because nothing in the standard tells them.
ERC-5805, the standard that governs delegated voting on Ethereum, describes delegation as a single assignment: an account delegates to itself, to a third party, or to address(0), and the delegatee's checkpointed voting power moves accordingly. The interface it specifies is delegates(account), delegate(delegatee), delegateBySig(...), getVotes and getPastVotes. There is exactly one expiry parameter in the whole standard and it belongs to delegateBySig, where it is the deadline on the signature, not on the delegation the signature creates. Nothing in the standard lapses, decays, or requires renewal.
That is a reasonable default for a token contract – the alternative is a contract that silently disenfranchises people – but it means a delegation made once during an airdrop claim in 2021 is still live in 2026 unless the holder returns and changes it. The delegate may have left the ecosystem, wound up their organisation, or lost the key. The weight stays where it was pointed.
Undelegated tokens are inert: they count toward supply and toward nothing else. Delegated tokens held by a delegate who never votes are worse than inert, because in any DAO whose quorum is a fraction of delegated supply they raise the bar that everyone else has to clear. The arithmetic of the three competing denominators is set out on voting power activation; the consequence is the one that matters here.
A delegate sitting on 2% of delegated supply and voting on nothing does not abstain neutrally. They add two points to the denominator of every quorum calculation while contributing nothing to the numerator, so the active electorate has to supply a strictly larger share of the votes to pass anything. Dormant delegation is a tax on the members who do show up, and it compounds quietly, because delegated weight is normally reported as a sign of engagement rather than as a liability.
It also distorts the standard defence of delegated governance. The claim that a DAO with 40% of supply delegated has a healthy electorate is only true if that 40% votes. Where it does not, the DAO has the worst of both: a hard quorum priced against the full delegated set, and a real turnout drawn from a fraction of it. See voter apathy for the participation side of the same gap.
The Optimism Collective's Operating Manual solves the quorum tax by refusing to count dormant weight in the first place. Its Token House quorum is 30% of the "total active votable OP supply" as of the start of the voting period, and it defines both halves of that phrase: votable supply is the OP that has been delegated and can therefore vote, and it "is considered active if delegated voting power has been utilized within the past 6 months." A delegate who goes six months without casting a vote does not merely fail to help reach quorum; their weight stops being part of the number quorum is measured against. Abstain votes count as utilisation, so the bar for staying in the denominator is participation of any kind rather than agreement.
The mechanism is honest about being a policy rather than an autonomous contract rule. The figure lives in a VotableSupplyOracle – a small Ownable contract that keeps a checkpointed history of the votable supply and is, in its own comment, "managed by Optimism Governance." The six-month test is computed off-chain and written to that oracle; the governor then reads the checkpoint. So the rule is enforced on-chain in the sense that quorum reads a number nobody can retroactively edit, and governed off-chain in the sense that a privileged owner decides what that number is. That is a real trust assumption and it is worth naming, but it buys something no stock OpenZeppelin Governor configuration offers: a quorum denominator that shrinks when delegates stop showing up.
The Arbitrum DAO takes the opposite approach: leave the denominator alone and buy the attendance. Its Rewarding Active Delegates (RAD) program pays enrolled delegates per proposal against a published monthly bar, and the bar moves with the workload. The June 2026 results state the rule exactly: four or fewer proposals in a month sets the participation threshold at 50%, five or more sets it at 75%. June 2026 was the first month with five governance actions in scope, so the threshold went back to 75% – at least four votes out of five.
The published outcome is the useful part, because it is a rare public count of how many funded delegates clear an explicit attendance bar. Across those five proposals, 36 enrolled delegates cast votes on 123 delegate–proposal instances, which the report scores as 68.33% program participation. Twenty delegates were paid, sharing $29,000; the other sixteen – named in the report, and including several well-known governance firms – received nothing for the month on grounds of insufficient participation, insufficient voting power, or zero eligible voting power. The program also prices the quality of the vote separately from the fact of it: a vote cast without a public rationale within five days keeps its eligibility but takes a 10% trim on that proposal's reward, and 66.67% of June's voted instances carried an on-time rationale.
Two structural weaknesses show up in the same records. The first is that paying for attendance measures attendance: a delegate can clear 75% by voting with the majority on five proposals they did not read, and the rationale trim is a 10% nudge rather than a real test. The second is budgetary. The quarterly review for Q3 2026 cut every per-proposal budget by 28–40% – a constitutional on-chain vote fell from $15,000 to $10,000, a temperature check from $5,000 to $3,000 – because the budgets are denominated in USD but paid in ARB, and ARB had fallen from around $0.21 at the program's launch to around $0.077.
An accountability mechanism funded by a governance token inherits that token's drawdowns, which is precisely when participation is hardest to sustain. The program as treasury spend – its price list, its published payroll corrections, and how it compares to paying for seats or paying in delegated weight – is covered on delegate incentive programs.
The third approach leaves both the quorum and the incentive alone and changes what a delegation is. Agora's Alligator contract, deployed for Optimism's governor, replaces the standard one-address assignment with a rules object. The SubdelegationRules struct carries seven fields: maxRedelegations, blocksBeforeVoteCloses, notValidBefore, notValidAfter, customRule, allowanceType and allowance. Each is checked when a vote is cast through the delegation chain, and the validation path reverts with a named error when one fails – NotValidAnymore past notValidAfter, NotValidYet before notValidBefore, TooManyRedelegations, TooEarly.
Two of those fields are direct answers to dormancy. notValidAfter gives delegation an expiry date, so a holder can lend weight for a season rather than forever and force a renewal decision. blocksBeforeVoteCloses is subtler and more interesting: a subdelegate constrained by it can only cast within the final stretch of the voting period, which makes a backup delegate expressible in the contract itself – someone who acts only if the primary delegate has not, and cannot pre-empt them otherwise. allowance with a partial allowanceType completes the set by letting one holder split weight across several delegates rather than betting the whole balance on one address remaining active.
The cost is complexity: rules are set per delegation by the holder, which means the mechanism only helps holders engaged enough to configure it – a smaller group than the one that delegated and stopped paying attention. Expressiveness at the contract layer does not by itself reach the dormant delegator.
Every answer so far asks something of the holder in advance – renew a delegation, pick a better delegate, configure a rules object. That is the limit named at the end of Answer 3: the mechanism reaches the engaged holder, and the problem is the disengaged one. A fourth approach inverts the timing and asks nothing until the holder actually disagrees.
GovernorCountingOverridable, added to OpenZeppelin Governor in v5.2.0 on 8 January 2025, lets a delegator cast an override on a proposal their delegate has already voted on. The delegate's tally is reduced by the overriding balance – the contract emits VoteReduced against the delegate and OverrideVoteCast for the holder – and each holder may override once, guarded by GovernorAlreadyOverriddenVote. Nothing has to be arranged beforehand: the delegation stands, the delegate keeps voting on everything else, and the holder intervenes only on the proposal they care about.
Two constraints bound how far this goes. The token has to inherit VotesExtended, which checkpoints delegations and balances rather than voting units alone, so a DAO already live on a plain ERC20Votes token cannot switch this on without changing the token – the same migration cost that has kept most Governor-family DAOs on the modules they launched with. And overrides can only be cast while the proposal is active, which the module's own source flags as a composition hazard: a governor that also runs early closure, such as GovernorSuperQuorum, can end the poll before the holders who disagreed have used the right the DAO just gave them.
Set against the other three, the trade is clear enough. Answers 1 and 2 act on the quorum and the delegate's incentives without touching the delegation. Answer 3 makes the delegation itself expressive but only for holders who configure it. Answer 4 is the only one that costs the holder nothing until the moment they want it – and the only one that requires the DAO to have chosen the right token contract years earlier.
ENS DAO is the live case where the problem is not one dormant delegate but a delegated set that keeps shrinking. In a July 2026 draft proposal, ENS contributor Alex Van de Sande argued that total delegated votes have been "consistently going down" apart from two jumps that were the same tokens being re-delegated, that turnout has fallen from proposals routinely drawing 3M votes to proposals struggling to make quorum, and that one delegate now holds enough weight to both meet quorum alone and outvote the next 50 delegates combined.
His diagnosis is a delegation cohort frozen at airdrop: tokens delegated once at claim, then sold to buyers who never delegate, so the electorate decays with every transfer while the supply stays constant. He notes that the DAO's own MetaGov stewards had already tried free re-delegation, distributing tokens to new contributors and paying holders to stake and delegate, "with mild successes" that did not reverse the trend.
The proposal's remedy – delegating 5M of the DAO's own ENS to stakeholders selected by metric rather than by election – is unresolved and contested, and it sits against a second live fight over whether a Foundation with a five-seat board should take on the DAO's operating functions. Both threads are worth reading as primary material on the same question: when delegated weight thins out, a DAO can respond by re-seeding the delegate set, by shrinking the denominator, or by moving the decisions somewhere delegates do not sit. ENS is publicly weighing all three.
They can make dormancy visible and costly. Optimism's six-month rule and Arbitrum's published payout tables both convert "this delegate stopped voting" from private knowledge into a number the DAO acts on. Neither existed in the original delegated-governance design, where the only remedy was a holder noticing and re-delegating by hand.
They cannot make a delegate read the proposal. Every mechanism above measures the observable act – a vote cast, a rationale posted within five days – because judgment is not measurable on-chain. Paying against attendance can even invert the goal, rewarding a delegate for voting on a question they had no view on rather than sitting out honestly.
They cannot fix a delegator who has stopped caring. Expiry, subdelegation rules and partial allowances all require the holder to act again. The population they most need to reach is defined by not acting.
They interact with the quorum design. A DAO that both prunes dormant weight from the denominator and enforces a high fixed quorum can find its quorum getting easier in the same month its electorate got smaller. See quorum and threshold design; the two settings must be read together, not chosen separately.
A caper has no delegation primitive at all. There is no delegate entry point in the contracts, so there is no delegate to go dormant, no delegated-supply denominator to inflate, and nothing to expire. Voting is first-person: the ballot is authorised by the voter's own account and requires a proof of at least one of that caper's tokens, and the weight it carries is the product of the voter's token balance and their balance of the caper's soulbound vote token, over the vote-token supply and the token's circulation. The vote token is minted only to the account that earned it – 1 per ballot cast and on nothing else, since trades stopped minting it on 11 September 2026 – and is soulbound thereafter, so the participation half of a member's weight cannot be lent, bought off another holder, or inherited from an inactive representative.
The consequence for dormancy is the sharp one: it is self-limiting rather than socialised. A member who stops voting stops accumulating the factor that gives their holdings weight, and their stake never sat in anyone else's column raising the bar for the members still voting. The same formula also governs the treasury exit – the contract requires a positive amount of both governance tokens and vote tokens, withdrawn from the member's own account, so a holder who has never cast a ballot has no participation-weighted claim on the treasury to exit with.
What this design gives up is the thing delegation exists for. A member with little time cannot lend their voice to someone who will study the proposal, and there is no professional delegate class to do the reading. Caper trades delegation's leverage for its accountability: nobody can represent you, and nobody can misrepresent you by doing nothing.
SubdelegationRules struct) and AlligatorOP.sol (rule validation and its named errors)expiry that is a signature deadline)VoteReduced/OverrideVoteCast, and the active-proposal and early-closure constraints) and VotesExtended.sol (the delegation and balance checkpoints it requires)