A DAO service provider is an outside firm paid by a DAO to run a defined part of its operation – indexing, treasury management, protocol engineering, growth, governance analytics. It is a distinct category from the two things it is usually confused with. A grantee is funded to build something of its own that the DAO would like to exist; a service provider is funded to deliver a service to the DAO, and the DAO is the customer. A contributor is a person the DAO compensates directly; a service provider is a firm with its own payroll, its own other clients (or a contractual promise not to have any), and its own commercial interest in being renewed.
That last property is what makes this a governance topic rather than a procurement footnote. A DAO that buys operating capability from a firm has created a counterparty with a durable incentive to keep the engagement alive, more information about the DAO's operations than most delegates have, and – in the mature cases – a hand in designing the process by which its own renewal is decided. The published cost of these engagements is covered on DAO governance operating costs. This page is about the relationship: how the firm is chosen, on what terms it is held, and what actually happens when the DAO wants out.
ENS DAO's Service Provider Program is the most fully specified version of this model, and its third season is instructive because the DAO wrote down the rules before running the process. The SPP3 authorization does four things in one vote: it authorizes a single cycle (renewal requires a new DAO vote, and the proposal states plainly that nothing in it commits the DAO to future seasons), it fixes the eligibility test (the primary deliverable must be a service to ENS; projects whose value accrues mostly elsewhere are ineligible), it names the committee that will execute the selection, and it caps the budget at 20% of trailing protocol revenue rather than at a fixed number.
Splitting authorization from selection is the design choice worth copying. SPP3 requires two votes from delegates: one to approve the programme, budget and committee, and a second to ratify the cohort the committee returns with. The stated reason is that the earlier season made delegates rate and rank every applicant themselves, which the proposal describes as a structural problem rather than a one-off inconvenience. Delegates are asked to judge a process and a panel, which they can do, instead of twenty-six commercial proposals, which they cannot.
The cohort recommendation shows the funnel that produced: 26 applications requesting a combined $12.2M against roughly $3.25M available; one disqualified at the eligibility gate; the remaining 25 scored against the published rubric with a structured interview for each qualifying team. The committee recommended four providers – Namespace ($500,000), Goldsky ($450,000), Unruggable ($400,000) and Fluidkey ($340,000), totalling $1,690,000 – and published its assessment of each, plus an aggregate explanation of why the rest were not funded. The committee's published per-criterion scores, and the question of what those numbers actually decided, are taken up on scoring and review gates. One committee member recused from evaluating and voting on a provider under the programme's conflict-of-interest rules, which is the minimum viable form of the disclosure this model needs.
The mechanics of the call that produces a cohort – the rubric, the deadline, the probity rules, and what an open call surfaces when nobody answers it – are treated separately in DAO procurement and requests for proposals.
Arbitrum DAO's arrangement with Entropy Advisors runs the opposite way: one firm, no cohort, and exclusivity written into the title of the renewal proposal. The firm was bootstrapped by an Arbitrum Foundation grant and works only for this DAO, which buys deep context at the price of having no comparator – there is no second bidder to price against and no obvious way to run a counterfactual.
What the DAO bought instead is an alignment and termination structure. The renewal's 15M ARB alignment allocation was revised mid-discussion into two parts: 5M ARB to the firm on a one-year cliff and three-year vest, and 10M ARB reserved for the oversight committee to negotiate against milestones, KPIs, equity or time-based bonuses rather than paid out on signature. The termination language was clarified in the same revision: if the DAO ends the relationship, all unearned ARB from that 10M returns to the treasury. Structurally this is the answer to exclusivity – if you cannot create competitive pressure at selection, create it at the vesting cliff.
A smaller DAO can reach the same place with a shorter clock instead of a vesting schedule. Beefy engages a single firm, Staworth, on the same sole-provider basis, but the engagement runs six months and then stops: renewal is its own public proposal, priced line by line, and the firm has to publish what it shipped to win the vote. Neither device creates a rival bidder. Both create a moment at which the DAO can decline.
Note also what the revision says about who the firm answers to. Once the DAO's operating company is sufficiently operationalized, it becomes the firm's counterparty and client, acting as a proxy for the DAO. The DAO stops being the direct commercial counterparty to its own strategy firm – which is model three.
The OpCo proposal makes the DAO's procurement problem explicit before solving it: a DAO is structurally slow at evaluating and engaging specialised contributors, initiatives often have no clear owner, and when a service provider disengages there is frequently nobody accountable for continuity. OpCo is a DAO-adjacent legal entity that hires staff and negotiates with service providers on the DAO's behalf – an operational layer that exists so the DAO does not have to be a counterparty to every engagement it depends on.
The interesting part is the drawdown ladder, because it is a governance mechanism disguised as a finance control. The proposal earmarked 30M ARB (explicitly subject to change between the Snapshot and Tally votes) to cover the first 30 months, of which 4M ARB forms a vesting, performance-linked bonus pool for internal employees and the oversight committee, and most of the remainder is liquidated until $12M in cash equivalents is reached, with the leftover ARB returned to the treasury. Against that balance: $2.5M released upfront, then up to $500K per month drawn freely, $500K–$1.5M requiring the oversight committee's approval, and anything above $1.5M requiring a Snapshot vote of the DAO. The entity is fast by default and slow exactly where the amounts get large enough to matter – the same graduated-threshold logic that appears in treasury management and in subDAO mandates.
Every procurement model looks sound in its authorizing proposal. The signal comes from the first time following it costs the DAO something. SPP3 supplies a clean instance. The committee's cohort originally had five selectees; the provider chosen for the marketplace and revenue vertical declined its award, leaving that vertical unfilled and $1,560,284 of authorized budget uncommitted.
The committee had an easy route available and said so publicly: go back to the applicants it had already scored and renegotiate scope privately. It declined to, on the grounds that SPP3 ran under one published rubric and one deadline that every applicant had agreed to, and that operating outside those bounds would not be in the programme's interest. Instead it proposed an open, tightly scoped RFP for the unfilled vertical, funded from money the DAO had already authorized: a single award of up to $500,000, milestone-gated, with roughly half streamed through go-live and half released against traction gates verifiable on-chain, and the balance returned to the treasury.
Three properties are worth naming, because they are what a DAO should look for in any procurement it approves. The remedy asked for no new money. It did not delay the four ratified providers. And it converted an awkward vacancy into another run of the same public process rather than a private conversation among people who already had the DAO's confidence.
The re-tender has since run to a conclusion, and the result argues for the model and complicates it in the same breath. Delegates authorized the RFP on 19 July 2026 by 1,191,366 to 156,420 with 66,837 abstaining. It drew nine applications requesting a combined $4,295,000 for the one award; one was returned at the eligibility gate and the other eight were interviewed and scored.
On 27 August the committee recommended Nomentum Labs, which operates the ENS marketplace Grails, for the full $500,000 – $90,000 paid up front in three monthly instalments, $310,000 opened as a stream only once the committee verifies an ENSv2-readiness milestone, and $100,000 held in four $25,000 gates released against attributed protocol revenue and distinct-wallet counts. The term co-terminates with the SPP3 cohort in Q3 2027 and anything unreleased returns to the treasury. The executable carried the DAO's on-chain governor 1,399,714 to nil with 64,350 abstaining against a 1,000,000 ENS quorum, and was executed from the timelock on 10 September 2026.
Hold the two halves of that outcome together. Nomentum Labs was formed by the three engineers who built and run Grails at EthID – the provider that was selected in the SPP3 cohort and then declined its award, creating the vacancy the RFP existed to fill. The product, brand, domain, repositories and infrastructure transfer to the new entity, and the committee met the double-funding question head on: EthID declined before the cohort executable passed and received no SPP3 funds, and Nomentum is a separate legal entity with no prior award. So the open re-tender ended by funding the same people on the same product a private renegotiation would have reached, at $500,000 rather than the $1,560,284 the vacancy freed.
What the process bought was not a different answer. It bought a published rubric, a stated award threshold of 3.0 weighted, a winning score of 3.84, a nearest applicant at 2.97 that anyone can see fell below the bar, and eight other teams that got to compete for the seat. That is the realistic case for following a procurement process at the moment following it is inconvenient: it does not promise a better selection, it makes the selection checkable. What that number did and did not decide is taken up on scoring and review gates.
Note the contrast with the software a DAO depends on, which is bought with a company card rather than a proposal and therefore has none of this process attached: DAO tooling discontinuity.
Caper has no service-provider layer, and that is a real answer rather than a gap being talked around. There is no delegate, steward, council or committee class anywhere in the contracts, and no streaming, vesting or recurring-payment primitive: a payment is a proposal option of the PAYOUT kind carrying a currency, an amount and a recipient account, and settlement of a passing proposal is the transfer.
The consequence for this topic is narrow and specific. A multi-season engagement cannot be expressed as one on-chain object, so it decomposes into a sequence of discrete proposals, each of which has to win its own vote. There is nothing to roll over and nothing to claw back – the money simply stops unless the members affirmatively send the next tranche. Renewal-by-default, the first failure mode above, is not a policy a caper has to adopt; it is unavailable. And a provider that accumulated a large position could not simply outvote the members on its own renewal, because vote weight multiplies holdings by an earned, non-transferable participation record, so a bag on its own carries no decisive weight. A member who disagrees with an engagement also has a second move beyond voting against it: a standing exit that returns their share of the treasury.
What Caper does not offer is a better selection process. There is no rubric, no committee, no RFP and no scoring – the DAOs above built those because choosing well among 26 applicants is a hard problem, and nothing in a shorter leash solves it. The honest summary is that Caper makes engagements easy to stop and offers nothing at all for making them easy to choose.
0x323A76393544d5ecca80cd6ef2A560C6a395b7E3 on 9 September 2026: 1,399,714 for, none against, 64,350 abstaining. Re-read on 5 October 2026, its state is Executed; the execution transaction is dated 10 September 2026.