Proposal spam is what happens when a DAO's proposal surface is cheap enough to post into that people who are not participating in governance start using it as a billboard. The posts are usually airdrop phishing: a title promising a claim, a body pointing at a wallet-draining site, published into the same list that carries the treasury votes. The DAO ignores them, the votes go on as normal, and nothing about how the organisation actually governs changes.

What changes is what can be measured. Almost every published governance statistic is a ratio taken over "proposals in this space": participation rate, average turnout, the share of votes that clear quorum, the trend in any of those over time. Spam contributes nothing to the numerator, because nobody votes on it, and everything to the denominator, because it is a proposal by the API's definition. The result is a set of numbers that look like evidence of decline and are an artefact of counting.

This page is about that measurement hazard rather than about phishing. The defences against posting are covered under proposal throttles and rate limits; what follows is what the spam does to anyone reading the resulting data, including the DAO itself.

A worked example, counted both ways

GnosisDAO governs on the Snapshot space gnosis.eth, which carried 257 proposals when it was paginated off Snapshot's public GraphQL API on 18 August 2026. 149 of them are closed and carry the space's 75,000 GNO quorum, which makes that set the natural denominator for any quorum question. 87 of those 149 are not governance. They are airdrop-phishing posts in about a dozen recurring templates, running from January 2024 to March 2026, and 26 of them drew literally zero votes.

Counted over	Proposals	Cleared 75,000 GNO	Median turnout
Everything the API returns	149	44 (30%)	1.02 GNO
Genuine governance only	62	44 (71%)	about 80,000 GNO

The numerator is the same under both counts, because no spam post has ever come near 75,000 GNO. The entire difference is a denominator inflated roughly 2.4 times. One version of this DAO fails to reach quorum on seven votes in ten and turns out a single token; the other clears quorum on seven in ten and turns out eighty thousand. Only one of them exists.

A median of 1.02 GNO is the tell, and it is the cheapest check available: if the middle proposal in your sample drew about one token of support, you are not measuring governance, you are measuring the spam. Any participation statistic taken off a public Snapshot space without a filter has this problem, and the direction of the error is always the same, because spam only ever adds silent rows.

The DAO tried to fix it and the fix missed quorum

GnosisDAO noticed. On 29 July 2024 it put GIP-109, “Should the GnosisDAO introduce Snapshot moderators to reduce spam?”, to a vote. It drew 72 voters and 3,942.06 GNO, of which 3,933.79 voted for, 0.004 against and 8.27 abstain. Support was effectively unanimous. It failed anyway, because 3,942 GNO is 5.3% of the 75,000 GNO quorum, and the spam kept arriving for another twenty months.

That is the sharpest form of the problem. The remedy for a hazard that degrades the quorum statistic had to clear the quorum, and a low-stakes housekeeping proposal is exactly the kind that does not. It is worth setting beside quorum and threshold design: a threshold calibrated for treasury decisions also governs the maintenance work, and maintenance rarely mobilises anybody. See voter apathy for the general shape.

A proposal threshold is not the same as a spam defence

The obvious response is to charge for posting. GnosisDAO already does. Read live on 18 August 2026, the space's validation is Snapshot's basic strategy with minScore: 1 and onlyMembers: false: an author needs one GNO of voting power, counted across mainnet, Gnosis Chain, beacon-chain deposits and delegation, and anyone clearing that bar may post. One GNO is not free. It did not work.

The reason is that a threshold prices the author, and a phishing campaign is priced against its take rather than against the DAO's dignity. A single qualifying address can post the same template twenty times, so the cost is paid once and amortised over the whole run. Defences that scale with the number of posts behave differently:

• Per-proposal deposits charge every post rather than every author, and can be refunded when a proposal is genuine, which is the shape most likely to actually bind.
• Sponsorship requires an existing participant to co-sign, moving the decision from an automatic score check to a person who can say no.
• Moderator roles, the GIP-109 answer, remove posts after the fact. They work and they are governance overhead, which is why they need a vote and why the vote can fail.
• A forum stage before the ballot keeps the on-chain or Snapshot surface for things that already survived a discussion, which is why DAOs running a mandatory proposal lifecycle tend to have cleaner spaces.

How to count a proposal surface honestly

The filter is the measurement. Two traps are worth naming because both produce a wrong answer that looks clean:

• Do not anchor a title regex. Real GnosisDAO proposals carry leading zero-width characters (​ GIP-134, ​ GIP-121) and prefixes such as [Redo] GIP-110, so a pattern anchored at the start of the title silently drops genuine governance into the spam pile. Match on a case-insensitive substring instead.
• Do not trust the pattern alone. Real proposals do not always carry a number: "Should Gnosis DAO conduct a $30 million buyback program?" drew 193 voters and 106,808 GNO under no GIP prefix at all. Spam also imitates the numbering, and two of the fakes in this space are titled GIP-92. The residual has to be read by hand.

Three habits follow. State the denominator whenever you publish a rate, because "44 of 62" and "44 of 149" are the same underlying fact and support opposite claims. Report the median turnout beside the mean, since a median near zero identifies a polluted sample instantly. And treat governance dashboards that count proposals per space, including DeepDAO and Boardroom, as reporting what the API returned rather than what the DAO did, unless they document a filter.

How Caper approaches this

A caper charges per proposal rather than per proposer, which is the shape that binds. create_proposal asserts payment.amount() == proposal_fee in XRD on every call, and the proposal fee is separately asserted to be at least the settlement-gas budget, 500 XRD, which is also the genesis default (contracts/logic/src/lib.rs). The author must additionally present a proof of holding at least one of that caper's own tokens. So opening a proposal costs the same on the twentieth attempt as the first, and the money is committed up front to the gas the proposal will consume when it settles.

That does not make a caper spam-proof, and it is not offered as a claim that it is. What it does is keep the denominator honest for the same reason: a list where every entry cost 500 XRD to create is a list you can count without a filter. The measurement problem on this page is downstream of a pricing choice, and pricing per post rather than per author is the version that survives someone deciding to post twenty times. See proposals for the full lifecycle.

References

• Snapshot, public GraphQL API. The 257-proposal census of gnosis.eth, the 149 closed proposals carrying the 75,000 GNO quorum, the spam classification and the space's validation and filters settings were all read from it on 18 August 2026.
• GnosisDAO, GIP-109: Should the GnosisDAO introduce Snapshot moderators to reduce spam? (29 July 2024): 72 voters, 3,942.06 GNO, 5.3% of quorum, closed without reaching it.
• Caper, contracts/logic/src/lib.rs: create_proposal (the XRD proposal-fee equality check and the one-token holding proof), SETTLEMENT_GAS_BUDGET and the genesis proposal_fee of 500 XRD.