A delegate incentive program is a standing budget a DAO votes to itself for the purpose of getting its own delegates to show up. It is one of the few governance interventions that shows up as a line item, which makes it unusually legible: unlike quorum design or ballot format, a delegate incentive program has a price, a payroll, and – in the better-run cases – a published monthly ledger naming who was paid and who was not.
This page is about those programs as spending. The neighbouring question – whether delegated weight should expire when a delegate goes quiet – is covered on delegate accountability and dormant delegates, and the supply-side arithmetic these programs sit on top of is on voting power activation. What follows is the money: what the three live designs actually purchase, what their own reports say they got, and the two failure modes the published data exposes.
The delegate payroll is also only one line in a DAO’s governance budget, and usually not the largest. What the councils, stewards, operating companies and service providers around it cost, and why those figures rarely compare across DAOs, is on DAO governance operating costs.
“Paying delegates” describes three structurally different transactions, and conflating them is the most common error in the discourse around them.
Cash per vote. The DAO sets a budget per proposal, splits it among delegates who met a participation bar, and pays in its own token at a fixed dollar value. This is the Arbitrum DAO’s Rewarding Active Delegates (RAD) design. What is bought is attendance on a specific list of votes.
Seats and roles. The DAO funds councils, boards and committees with per-seat stipends, and the delegate incentive is a job rather than a per-vote rate. This is how the Optimism Collective spends the bulk of its governance budget. What is bought is a person’s ongoing attention to one domain.
Delegated weight. The DAO – or another DAO holding its token – pays nothing at all, and instead routes voting power to chosen delegates. What is bought is influence for the recipient and, the payer hopes, competence for the electorate. Uniswap DAO’s Arbitrum delegate program and the July 2026 ENS draft below are both of this kind.
The three have different failure modes, and only the first has a real budget line. That asymmetry matters: the programs that are easiest to criticise are the ones whose costs are visible.
Arbitrum’s RAD program prices participation by proposal category, with a pot per proposal and a cap per delegate. The quarterly budget review published on 1 July 2026 states both the old and the new schedule, effective for proposals starting in July:
| Proposal category | Budget (old → new) | Per-delegate cap (old → new) |
| On-chain constitutional quorum | $15,000 → $10,000 | $700 → $550 |
| On-chain non-constitutional quorum | $7,000 → $5,000 | $500 → $400 |
| Off-chain decision-making (non-constitutional) | $7,000 → $5,000 | $500 → $400 |
| Off-chain election | $7,000 → $5,000 | $500 → $400 |
| Off-chain temperature check (non-binding) | $5,000 → $3,000 | $300 → $250 |
Two properties of this design are worth naming. First, the DAO’s governance payroll is a function of how many proposals it runs, not of how hard the decisions were: a quiet month is cheap and a busy month is expensive regardless of what was at stake. Second, the eligibility bar moves with the workload rather than being fixed – the June 2026 results state that four or fewer proposals in a month sets the participation threshold at 50% and five or more sets it at 75%, and a missing vote rationale trims a delegate’s payout by 10%.
Most other programmes pay a fixed stipend to a short list of delegates instead of a price per vote. Uniswap's Delegate Reward Initiative, proposed by StableLab in February 2024 and running since June, pays each eligible delegate who meets its participation requirements up to $6,000 of UNI a month (The Block); its third cycle requested $540,000 of UNI to pay 15 delegates for six months (Uniswap proposal 80).
Aave's Orbit programme, started by the Aave Chan Initiative in mid-2023, pays recognised delegates in GHO and refunds the gas they spend voting (Aave governance), and was renewed for the first half of 2026 (Aave governance). Lido's pilot paid $150,000 of LDO between 19 September 2024 and 18 March 2025, in equal shares to delegates holding at least 2m LDO in delegations, capped at $15,000 per delegate per quarter and reviewed by a delegate oversight committee; Lido later extended it (Lido governance; Boardroom).
RAD budgets are denominated in US dollars and paid in ARB. The July 2026 review is explicit about what that did. When the program began ARB traded around $0.21; at the time of the review it sat near $0.077, a fall of roughly 65%. Because the dollar budget per vote never changed, “each vote has been drawing two to three times more ARB from the treasury than it did at launch, even though the dollar value delegates receive is the same.”
The program’s own figures make the drift concrete: an on-chain constitutional vote had cost ~131,300 ARB per proposal on average across the program’s history, but ~200,000 ARB at the old budget and the prevailing ~$0.075 price; the new $10,000 budget brings it back to ~133,300 ARB, close to the historical norm.
The general result is that a governance payroll set in fiat and paid in the governance token is procyclical against the treasury. The token cost of the same nominal participation rises exactly as the token falls – that is, the DAO spends the most tokens precisely when its treasury is worth the least and its runway is shortest. Every DAO paying contributors in its own token at a dollar rate carries this exposure; delegate programs simply make it easy to see, because the ledger is public and monthly. Note also who absorbed the correction: Arbitrum restored the ARB spend by cutting the delegates’ dollar income 28–40%, so the price risk was ultimately borne by the delegates, not by the treasury that had been over-spending in token terms.
In May 2026 the RAD program published a correction and post-mortem for its April report, and it is the most instructive artefact any of these programs has produced. A bug in the monthly automation that lists Snapshot proposals closing in the reporting month silently dropped one of them – the vote on updating the Code of Conduct and DAO Procedures. The first report was therefore built from three votes rather than four.
What changed and what did not is the point. The corrected build used the same 34 enrolled delegates and produced the same headline figure: 88 “Yes” votes over 136 slots (34 delegates × 4 proposals) is 64.71% program participation, exactly the number the flawed three-vote report had published. But the budget envelope moved from an implied $17,000 to $24,000, and rewards owed moved from $14,500 to $20,000 – $5,500 more than had been paid out on 5 May.
The reason is structural rather than accidental: participation is a ratio, so it is invariant to the number of proposals in the denominator, while the payroll is per proposal, so it is not. A published participation rate can therefore be perfectly correct while the payment ledger built from the same dataset is materially wrong. Any DAO that treats a participation dashboard as evidence that its delegate payroll reconciles has confused a normalised metric for an audit. The RAD program’s own bi-annual transparency reporting is what surfaced the delta at all.
The same post-mortem exposes a second, subtler property. With four proposals in scope, a delegate’s monthly participation can only take five values – 0%, 25%, 50%, 75% or 100% – and in April’s corrected data the observed distribution was 19 delegates at 100%, six at 50%, nine at 0%, and zero at 75%. A rule written as “at least 75% participation” is therefore not a fixed standard: on a four-vote month it is operationally a demand for a perfect record, while on a twelve-vote month it forgives three misses. The strictness of the rule changes month to month, without anyone proposing an amendment, purely as a function of how many proposals the DAO happened to run. This is the same granularity trap that afflicts percentage quorums on small electorates, discussed on quorum and threshold design.
The Optimism Collective spends on governance differently. Its Token House funds councils and boards with operating budgets rather than paying a rate per vote, and its delegate-facing incentive is retroactive: the Anticapture Commission has no operating budget at all, but active participation there – the threshold cited is voting on at least 50% – may count toward Retro Governance rewards.
A Season 7 participation and incentives report compiled by the delegate-research team SEEDGov puts Token House governance operating costs at 1,055k OP, up 38% from 762k OP in Season 6. The report is careful about why, and the caveat is worth carrying: part of the rise is that the Security Council’s 295k OP budget became a public Token House line for the first time, having previously sat inside Foundation expenses. A governance-cost series can jump because spending rose or because disclosure improved, and only the underlying report distinguishes the two – which is a standing reason to be sceptical of cross-season DAO cost comparisons quoted without their source.
On what the spend bought, the same report finds average turnout among the top 100 delegates at 49 of 100 per vote, up 6.8% on Season 6, alongside 22% fewer vote rationales and 25% less forum feedback than the previous season. Its own summary of the pattern is that voting happens and discussion does not. That is the recurring hazard of paying against an observable: attendance is measurable and cheap to satisfy, deliberation is neither, so a program that prices the first without pricing the second can raise turnout and lower the quality of the argument at the same time.
The third design spends no treasury cash. In October 2023, after Arbitrum’s ecosystem airdrop sent Uniswap a little over 4 million ARB, Uniswap governance voted to route part of that allocation through a Uniswap-Arbitrum Delegate Program rather than sell or idle it – described by its facilitators as “the Uniswap community’s unique entrance into metagovernance.” The delegation facilitators publish per-vote rationale on both DAOs’ forums and disclose conflicts where their own organisations hold separate Arbitrum delegations. The compensation here is influence, and the payer is a different DAO than the one being governed.
ENS is the case where paying cash was tried first and reported as insufficient. A July 2026 draft proposal by ENS contributor Alex Van de Sande argues that delegated votes and turnout in ENS DAO have both been falling, and lists what has already been attempted: making redelegation free, distributing more ENS to new contributors, and “paying users to stake and delegate their ENS.” His assessment is that these “have had some mild successes” but “clearly have not been able to reverse the overall trend.” The draft’s alternative is to stop paying for delegation and start allocating it – delegating 5M ENS from the DAO’s own majority holding to named stakeholder classes. It is a draft rather than policy, and should be read as one, but it is a rare on-the-record verdict from inside a major DAO on whether cash incentives fixed its delegation decay.
A fourth design pays the token holder for delegating rather than the delegate for voting. ZKsync distributed 35m ZK over two three-month seasons to holders who staked and delegated, 10m ZK in the first season against a target of 400m staked and 25m in the second against 1bn, at up to 10% a year, on staking contracts built by Tally (Tally). Arbitrum voted for a similar scheme on 9 September 2024: ARB staking would stream DAO revenue such as sequencer fees to holders delegated to an active participant, and it passed with 134.45m ARB for and 4.5m against (Tally). The working groups the vote created recommended deploying the contracts first without rewards (Arbitrum forum).
Obol pays delegates, but flattens the pay. From September 2025 Tally distributed 165,000 OBOL over six months to delegates whose Delegate Reputation Score, computed by Curia, averaged 65 or more over the previous five proposals. Each eligible delegate's share follows the square root of their delegated voting power rather than the power itself, so a smaller active delegate earns more per unit of weight than a large one (Tally; Obol OIP-4).
The strongest evidence on the limits of these programs comes from Arbitrum, which runs the best-documented one. On 1 June 2026, the research firm Blockworks – by its own description “the second biggest delegate in Arbitrum” and a contributor of research through the ARDC on incentive-program design, STIP and LTIPP performance, sequencer revenue and Timeboost – announced it was winding down its active delegate role. The stated reason was alignment with the firm’s own business goals, not dissatisfaction with the DAO, and the post encourages delegators to redelegate to contributors with the bandwidth to participate consistently.
A per-vote incentive is priced against the marginal cost of casting a vote. It is not priced against the opportunity cost of a research firm’s entire governance practice, and it cannot be: no plausible per-proposal cap competes with a business decision of that size. So the honest summary of what these programs achieve is narrow but real – they raise measured attendance among delegates who have already decided to be there, they make absence a published number rather than private knowledge, and they do not retain the largest professional delegates when the economics of those delegates change. On the general question of whether paying for turnout addresses the underlying condition, see voter apathy and governance participation; on delegate pay as one category of DAO payroll among several, see contributor compensation in DAOs.
A caper has no delegate to pay. Searching the three contract crates for any delegation primitive returns nothing – there is no delegate entry point, no delegated-supply figure, and therefore no professional delegate class for a budget to subsidise or retain.
Participation is also priced in the opposite direction. Voting is a paid-in act, not a paid-out one: vote() in contracts/logic/src/lib.rs asserts that the payment is XRD and is exactly the configured vote fee, then forwards the bucket in one fused hop to vote_apply in contracts/core/src/caper_dao.rs, whose first statement deposits it into the caper’s own treasury vault. There is no separate ballot crate and no vault escrow: a contracts/ballot package existed during the escrow-voting era and was deleted when that design was replaced, and the reinstated two-phase governance runs entirely in the logic and core tiers.
Both rejection paths abort the whole transaction rather than banking the fee – a ballot arriving outside the voting window fails on "Voting window closed", and a caller holding less than one governance token fails on "Must hold at least 1 DAO Token to vote". Balances are read live off the voter’s own account, so nothing is staked or custodied in order to vote. Every ballot that lands is a transfer from the voter to the treasury the vote governs, and it mints the voter VOTE_MINT = one soulbound vote token. See proposals for the full fee schedule.
What replaces the stipend is the exit claim. exit() asserts that the member holds both vote tokens and governance tokens before it will compute anything, and then pays the same weight the ballot carried – (t × v) / (V × T), the member’s token balance times their vote-token balance, over the vote-token supply times the token’s circulation. A member who has never cast a ballot – the only act that mints v since trades stopped minting it on 11 September 2026 – has no participation-weighted claim on the treasury at all. So the reward for showing up is retroactive, funded by the same treasury the votes decide, and proportional to a record the member paid to build – rather than a monthly payroll that has to be budgeted, priced in a falling token, and reconciled.
The cost of that design should be stated plainly. There is no way for a caper to fund a full-time analyst to read its proposals, no way for a member short of XRD to have their fee subsidised, and no way to lend voting weight to someone with more time. Caper trades the professional delegate layer away entirely rather than trying to pay for it well.