A review gate is the step where a DAO hands a funding decision to a small group of named people, asks them to score applications against a written rubric, and publishes the numbers. It exists because token voting answers the wrong question. A vote measures how popular a proposal is among people holding the token; it does not measure whether the engineering is sound, whether the team has shipped before, or whether the science is any good. Nearly every DAO that funds work at scale has arrived at the same patch: put competence in front of the ballot, and write down how it scored.
This page is about the number that comes out. It is a narrower object than the two topics it sits between. Procurement and RFPs covers how the call is run and who is eligible to answer it; DAO service providers covers who the DAO ends up contracting with and on what terms. Between the two sits a published score, and the question worth asking of any DAO that produces one is simple: what did it decide?
Three programmes published enough to answer that. In all three the honest answer is that the score decided less than the artifact implies.
The Optimism Grants Council reviews applications to the Governance Fund and publishes the results per cycle. Cycle 35's preliminary review results, posted 28 March 2025, are unusually legible: a single table listing every application, the OP requested, the average reviewer score, and the outcome. The post states the rule in one line above the table, that the score required to advance was 7 or higher, and reports that 13 of the applications passed preliminary review.
The table does not obey the line. Three applications advanced below the stated mark: Giza at 6.66, DeSyn Protocol at 6.00 and SwapBased at 6.00. In the same table, Gamma Strategies was declined at 6.66, Orderly at 6.33 and yUSD at 6.00. An identical 6.66 produced both outcomes in the same cycle, and a 6.00 both passed and was declined.
The explanation is in the budget line rather than in any inconsistency of judgement. The post records that finalists requested 15,365,599 OP against a Grants Council budget of 3,285,070 OP, an oversubscription of roughly 4.7 to 1. A pass mark cannot bind when the money runs out at a quarter of the qualifying field. Once that is true the score has quietly changed job: it is a ranking used to sort a queue, not a threshold that admits anyone who clears it, and the two are different instruments even though they are printed in the same column. The council's own framing points the same way, since the same post ties every award to the season's success metrics for Superchain TVL rather than to the number in the table.
Optimism's wider funding history is covered on Optimism Collective and, for the retroactive rounds that ran alongside the council, on retroactive public goods funding.
ENS DAO reached the same place from the opposite direction, and wrote it down. The DAO's Service Provider Program had asked delegates to rate and rank every applicant themselves. A February 2026 proposal argued that this had failed on three counts: delegates were hit with a barrage of lobbying, applicants spent weeks chasing up to a hundred delegates with the well‑connected ones advantaged, and most delegates did not know enough about the ecosystem to read the applications properly. The remedy was to stop voting on applicants and vote on a panel instead.
Delegates ratified exactly that. SPP3's program authorization passed on Snapshot in May 2026 with 1,478,297 votes for against 18,438 opposed, authorizing one cycle, naming the committee, and capping the budget at 20% of trailing protocol revenue. The committee then published its rubric before the submission window opened: four weighted criteria scored 1 to 5 in half‑point steps, with prior delivery history at 25%, scope clarity at 15%, milestone structure at 15%, adoption and ecosystem utility at 40%, and 5% of discretion for team and budget fit.
The sentence that matters sits directly under the weights table. The committee wrote that the rubric is an evaluation tool, not a selection formula, and that the final cohort would account for budget constraints, provider overlap and strategic gaps across the cohort as a whole. That is Optimism's Cycle 35 table stated as policy in advance rather than left for a reader to infer.
The cohort recommendation in July 2026 bore it out. Twenty‑six applications requested a combined $12.2M; one failed the eligibility screen and 25 were scored. Four providers were funded for $1,690,000 against roughly $3.25M available, and their weighted scores were published: Namespace 3.43, Goldsky 3.37, Unruggable 3.30 and Fluidkey 3.11, each out of 4.75 once the discretionary component is excluded. The whole funded cohort therefore sits in the rubric's adequate band, and the highest single criterion score anywhere in the published table is a 4.5 for prior delivery. The committee also stated plainly that several declined teams scored competitively and that redundancy rather than quality was the deciding factor, because the applicant pool contained multiple marketplaces, subname providers and indexing proposals and funding duplicative work reduces total programme value.
Two structural facts round it out. The committee left roughly $1.56M of authorized budget uncommitted rather than fund a marginal application, which is the strongest available evidence that the panel was exercising judgement rather than spending a pot. And the programme carries a $200,000 floor on requests, so the gate excludes small applicants before any scoring happens. The committee named that as a finding of the cycle, reporting demonstrated demand for discrete sub‑$200K funding that no active programme serves, and recommended the DAO stand one up.
ENS ran the same instrument again in 2026 under conditions that removed both of the confounds above, which makes it the nearest thing this topic has to a controlled case. The Marketplace RFP, authorized in July after an SPP3 selectee declined its award, funds exactly one provider out of a ceiling rather than a target, and states in terms that if no proposal clears the published bar then nothing is funded and the money returns to the DAO treasury. There is no cohort to shape and no budget that binds before the rubric does.
The committee published a marketplace-specific rubric weighted 25% prior delivery, 20% scope and milestone credibility, 35% revenue and adoption, 10% ENS and ENSv2 readiness, and 10% post-award self-sustainability – together with something SPP3 never stated, an award threshold of 3.0 weighted. Nine teams applied for a combined $4,295,000 against the single award, one was returned at the eligibility gate and eight were scored. The recommendation put Nomentum Labs at 3.84 weighted, top-scored by every member individually, with a per-criterion table running from 4.38 on prior delivery down to 2.62 on sustainability. It also published a number the cohort round never had to: the nearest applicant scored 2.97.
So here the score and the outcome coincide and the threshold looks like it did real work. Read it again and it did less than that. A pass mark that admits exactly one of eight applications has not been tested as a pass mark – it selected the team a plain ranking would have selected, and the two instruments only come apart when more than one candidate clears the line. What the 3.0 genuinely decided was the counterfactual: it is the reason the committee could have funded nobody and sent $500,000 back to the treasury, which is a discipline a ranking cannot express. The published runner-up score is what makes that claim checkable.
Note also which criterion scored lowest and what it weighed. Nomentum's 2.62 on post-award self-sustainability sits beside the committee's own written finding that the team acknowledged the DAO is its business model today, its subscription tier unbuilt and unpriced. That criterion carries 10% of the weight. The property the programme says it is buying – a provider that will not need the programme – is the one dimension the number is least able to move. The relationship the award actually creates is the subject of DAO service providers.
VitaDAO runs the same instrument on research rather than infrastructure, and supplies the failure mode neither of the others reaches. VDP-164, proposed in September 2025, asked for a $500,000 recoverable loan to take a therapeutic through non‑GLP toxicology toward an IND submission. Four reviewers scored it across seven categories for an overall conviction score of 2.3, against an average of 3.7 for the projects VitaDAO had previously funded.
Nothing then rejected it. The constitution's escalation ladder says a request of that size needs a forum poll and a Snapshot vote, and the proposal has never appeared on VitaDAO's Snapshot space. It did not lose a vote; the vote was never called. A scored gate that sits in front of the ballot can therefore produce a third outcome beyond pass and fail, which is that a proposal simply stops, with the record of what happened living in a forum thread rather than in a tally anyone can point to.
The durability of that record is not guaranteed either: gov.vitadao.com stopped resolving for at least a fortnight in August 2026 and was serving the thread again on 22 August. A gate that strands proposals leaves its only record on the least durable layer of a DAO's stack – the detail is on VitaDAO. The wider mechanics of research funding are covered on decentralized science funding.
Read together, the three programmes point at the same conclusion from different angles. The score is an accountability artifact, not a decision rule.
None of this is an argument against review gates. Every alternative on offer is worse: delegates who admit they cannot read 26 commercial proposals, or a token vote that measures popularity and calls it merit. The argument is against reading the number as though it were the decision. A DAO that publishes scores and lets its members infer that the top scores were funded is claiming a legibility it does not have, and ENS's contribution is to have said so in the rubric itself.
Part of the scoring is now machine-generated. Karma, the platform Optimism's Grants Council moved its applications onto in 2025, scores applications with a model and flags risk from an applicant's record in other ecosystems, leaving the decision with the reviewers (Karma; Karma). That does not change what this page measures. A generated score is still a number beside an application, and the gate is still wherever the budget and the committee decide it is; what changes is how cheap it becomes to produce one, and how much weight a reader may put on a figure no named reviewer wrote.
There is a consistent asymmetry in what these processes publish, and it points away from the people with the strongest interest in the answer.
ENS published a per‑criterion score table for the four providers it funded. Declined applicants received aggregate rationale grouped into recurring patterns, and the individual scoring records remain internal working documents available to the accountability body or the foundation on request. Optimism is the more transparent of the two on this axis, publishing a score for every application including the declines, though only a single average rather than the reviewers' individual marks. VitaDAO published a category‑by‑category digest for its declined proposal but not the reviewer document, and the loudest objection came from the chief executive of the company under review, who argued in the forum thread that the community should be able to read the reviews rather than rely on anonymous reviewers whose credentials nobody could check.
The pattern is that the applicant who won learns the most about why. This is understandable in each case and unfortunate in aggregate: the decline is where the information is, both for the applicant deciding whether to reapply and for a delegate trying to judge whether the panel it deputized is any good. A programme that discloses only its winners is asking to be trusted on precisely the decisions it is not showing.
The Marketplace RFP moved one step the other way without closing the gap: its rules required the winning score to be published, and the recommendation disclosed the nearest applicant's 2.97 as well, so a reader can at least size the distance between first and second. The eight declined teams still received grouped rationale and no number of their own. The wider failure of a committee whose work the DAO cannot check is covered on council dissolution and how DAOs fail.
Caper has no review gate, and no way to build one. There is no reviewer, panel, committee or steward class anywhere in the contracts, no rubric, no score and no eligibility screen. On the merits of choosing between 26 applications that is a gap rather than an answer, and it should be said plainly: nothing in Caper helps a group of people work out which of several proposals is worth funding. The DAOs above built rubrics because that problem is genuinely hard, and a shorter leash does not solve it.
What Caper does remove is the third outcome. A proposal carries an execution window on a clock, and once that window opens finalize_proposal is a public method: any account can call it, freeze the result and settle the proposal, and the winning payout is then executed through an equally public method. The party inconvenienced by the outcome is not the party who has to call the vote. A funding request on a caper cannot spend eleven months between a review and a ballot nobody schedules, because scheduling is not a privilege anybody holds.
The other structural difference is who the gate answers to. Vote weight in a caper multiplies a member's holdings by their earned, non‑transferable participation record, normalized against the caper's supply of each, so a large position on its own does not carry a decision. There is no separate body whose approval a proposal needs before members can vote on it, which means there is also no body a well‑funded applicant can lobby instead of the membership. And a member who loses the argument retains a standing exit that returns their share of the treasury rather than only the option to keep arguing. See what is a caper for the full model.