A company that stops has a procedure waiting for it: a liquidator is appointed, creditors are ranked, assets are sold, the entity is struck off a register. A DAO has none of that. There is no court with jurisdiction over a Snapshot space, no registrar to notify, and in most cases no legal person to dissolve. What a DAO has instead is the same instrument it has for everything else: a proposal. Winding down is therefore a governance action – and like every governance action, it only reaches as far as the contracts it can actually call.
That limit is the whole subject of this page. A dissolution vote can stop new spending and authorise a redemption. It cannot reach a position locked in someone else's contract, cannot un-delegate the voting weight the DAO accumulated elsewhere, and cannot make anyone keep paying for the forum where the vote was debated. The result is that a wound-down DAO does not disappear; it decomposes, in a predictable order, leaving parts of itself running for years.
The clean case looks like ordinary governance. Aura Finance, the vote aggregator built on top of Balancer's vote-escrow system, ended itself with AIP-78, “Orderly Protocol Wind-Down and Treasury Redemption for Aura”. It opened 9 April 2026 and closed on the 12th, carrying 17,171,688.83 vlAURA for, 0 against, 0 abstain, across eight voting addresses. It is the last proposal the aurafinance.eth space has ever recorded.
The shape recurs. Swapr, the DEX that outlived DXdao, shut its token down by the same route: “Shutdown SWPR token” opened 28 January 2025 and closed on 3 February with 4,935,266.20 for, 0 against, 0 abstain on 74 voters. Two dissolutions, two unanimous results, and not a single dissenting vote between them. PoolTogether ended its governance the same way in February 2026: 46 of 47 voting addresses for, with a single multisig casting more than three times the quorum on its own.
The unanimity is not consensus so much as self-selection. By the time a wind-down reaches a ballot, the holders who disagreed have usually already sold; the ones left are the ones who want the redemption processed. This is the mirror image of the pathology catalogued under voter apathy – a turnout collapse that helps the vote pass, because everyone still present wants the same thing. It also means the vote tally is close to worthless as evidence that a wind-down was well designed. Eight addresses agreeing is not a mandate; it is a quorum of the remaining interested parties.
CityDAO shows what a dissolution looks like when it is not unanimous. The collective formed in 2021 as a Wyoming DAO LLC, bought 40 acres of land in Wyoming, and was governed by holders of citizenship NFTs through its Snapshot space (Snapshot Labs). It ended with CIP 212, “Return Funds to Citizens”, which closed on 14 May 2024 with 92.6% of the voting weight for and 7.4% against, across 308 voters.
The proposer, one of the two contributors who had kept the LLC compliant, paid its taxes and run its bank accounts, gave the reasons: seven or more proposals had failed to pass, members could not agree how to spend the money, and millions of dollars left in a multisig whose signers kept changing was a risk. A follow-up, CIP 216, recalled unspent money from two sponsored projects into the same pool. DAO Times reports that $3.076m was returned to 4,808 addresses from June 2024, and that the 40-acre parcel stayed with CityDAO's LLC with no decision about what to do with it (DAO Times).
This is the part most wind-down write-ups miss, and it is checkable on-chain. Aura's entire product was a locked position: its VoterProxy contract 0xaF52695E1bB01A16D33D7194C28C42b10e0Dbec2 held Balancer LP tokens escrowed in veBAL. AIP-78 dissolved the protocol. It did not, and could not, break the lock.
Read against Ethereum mainnet at block 25,758,271 (15 August 2026), four months after the dissolution vote, the proxy still held 3,684,354.766 BPT locked in veBAL with locked__end = 29 April 2027, giving it a veBAL balance of 2,592,008.185 against a total supply of 3,573,920.103 – 72.53% of the entire vote-escrow electorate of a protocol it no longer participates in. Meanwhile the constituency that used to direct that bloc had evaporated: vlAURA outstanding stood at 332,317.70 against an 86,413,430.86 AURA supply, or 0.38%.
So Balancer inherited a governance problem it did not create: a supermajority of its escrowed voting weight belonging to a dead counterparty, immovable until 2027, and – critically – still counted in the denominator when measuring turnout. The DAO's answer is BIP-924, “Exclude Aura from Governance and Reduce Quorum”, filed 14 August 2026 and open to 18 August. Its remedy is two-part: overwrite the VoterProxy to zero voting power in the aggregator, and halve the quorum from 10,000,000 to 5,000,000 because the old bar was calibrated around a participant that no longer exists.
The two proposals immediately before it are the evidence for that arithmetic. BIP-922 and BIP-923 both closed 4 August 2026 unanimously – on eleven and ten voting addresses carrying 6,473,256.62 and 6,454,723.41 respectively, both short of the 10,000,000 quorum. BIP-924 itself, checked 16 August 2026, sits at 21,837,199.55 unanimous across five addresses, comfortably over the bar it proposes to lower. A DAO that cannot reach quorum has not lost its members; it has lost the one member who used to carry the quorum single-handed, which is the failure mode quorum design is least often stress-tested against.
A wound-down DAO decomposes in a consistent sequence, and knowing it tells a reader where to look for the truth months later. The parts that die first are the ones somebody has to keep paying a bill for.
forum.aura.finance nor docs.aura.finance had an A record at all. The forum thread is the discussion link on AIP-78 itself – so the deliberation behind the wind-down no longer answers at the address the vote points to. It survives only in the Wayback capture of 8 May 2026.app.aura.finance still served one at the same check.eth_call.DXdao makes the same point from the opposite direction, and adds a wrinkle worth internalising: a collective's domain is rented; its ENS name is not. DXdao's DNS-hosted forum, daotalk.org, today serves a gambling-affiliate site – and answers HTTP 200 while doing it. The DAOstack client it used, alchemy.daostack.io, has no DNS. But the site DXdao published to ENS and IPFS outlived both: dxdao.eth.limo served the original “The First Super-Scalable Collective” page, unchanged, years after the collective stopped, and the Wayback Machine captured it on 19 October 2026. Since late October 2026 the eth.limo gateway answers that name with a “Service Issue” error page.
The practical rule for anyone citing a wound-down org: archive the forum thread before you cite it, prefer the Snapshot space and the contracts for anything load-bearing, and try <name>.eth.limo before concluding the web presence is gone. A live DNS domain is the least trustworthy artefact in the set, not the most – the same lesson tooling discontinuity teaches about the software a DAO runs on.
A DAO that sunsets deliberately usually publishes a retrospective. What it rarely does is take down everything that contradicted it, and the result is a corpus that tells two incompatible stories depending on which page a reader lands on.
Coordinape is the cleanest specimen. Its memorial at coordinape.com is an honest, past-tense chronicle – it names the limits of the mechanism outright, and reports 50M+ GIVE allocated through circles, 500+ organizations at peak, and 64,000+ attestations on Base. Its own subpage for CoDAO, the $CO holder community named as steward of the IP, was never taken down: it remains frozen in pre-sunset future tense, still saying “Launch Coordinape” and describing a handover the memorial dates to the past. And the destination the memorial links to, codao.xyz, is now a 114-byte shell that redirects to a GoDaddy aftermarket listing – the steward DAO's own domain is for sale.
The generalisable probe is cheap: for any org you believe is closed, fetch every first-party host it cites and grep for future-tense roadmap verbs and live-app calls to action. A hit means the citation is anchored to an unmaintained artefact. First-party is an authority claim, never a freshness claim – a distinction that matters as much for a live project's support KB as for a dead one's landing page.
Most DAOs never hold a dissolution vote. They stop having proposals, and the secondary literature settles on an approximate obituary date that nobody checks. The ledger disagrees more often than not.
DAOstack is the standing example. Write-ups place its wind-down “around 2021”. Read via Blockscout and eth_call, the GEN token (0x543Ff227F64Aa17eA132Bf9886cAb5DB55DCAddf) still reports a total supply of exactly 60,000,000 across 7,251 holders, and the last proposal the GenesisProtocol mechanism ever processed was created on 17 January 2024 – nearly three years after the accepted date. Its record is a fitting epitaph: staked and boosted within a day, two votes cast in favour, and then no execution event at all. The organization it was raised for was DXdao's own avatar. The mechanism's last act was a proposal that passed nothing.
DXdao's own ledger tells the ending more bluntly than any post-mortem: of 148,976.876 DXD in existence, the single largest holder is the burn address, at 125,667.619 DXD or 84.35% of supply. The members redeemed and burned their way out one at a time; there was never a vote that said “we are finished”. This is the quiet death catalogued under how DAOs fail, and its distinguishing feature is that the treasury and the token keep existing long after the organization has stopped deciding anything.
Krause House shows the same death in its turnout figures. It formed in 2021 to buy an NBA team with its members' money, raised 1,000 ETH in an NFT sale, and bought the Ball Hogs, a team in the BIG3 league (Snapshot Labs). Its Snapshot space records 214 proposals. The median proposal drew 72 voters in 2021, 23 in 2022, 15 in 2023, 7 in 2024 and 5 in 2025, while the number of proposals fell from 120 in 2022 to three in 2025 (read 19 September 2026). The last, on 3 December 2025, asked the DAO to reimburse $2,000–$3,000 for a junk removal service to take its basketball court out of storage, because a member had been paying the storage bill personally; four addresses voted, all in favour (Snapshot). None of the 214 proposals asks the DAO to dissolve.
LabDAO is the limiting case of the same pattern, and it shows that the missing dissolution vote is not always the strongest evidence available. The DeSci tooling collective registered a Snapshot space on 22 December 2021 and, read keylessly against hub.snapshot.org/graphql on 17 August 2026, never ran a single proposal through it. Its own charter defined members as holders of a $LAB token, alongside a parenthetical conceding that no such token existed. There is consequently no last proposal to date the death by – only a website that went from live on 7 August 2024 to a parking page selling the domain for $1,890, and an organization profile quietly rewritten into the past tense. The full reconstruction is on the LabDAO page.
The substance of an orderly dissolution is the redemption: a stated formula, a stated pool, and a window in which a holder can claim. Swapr's shutdown published all three – 515,000 ARB liquidated into roughly $360,000, distributed at about $0.015 per SWPR across 23,988,449 circulating tokens on two chains. A holder could compute their claim from the proposal itself.
What separates that from the alternatives is not generosity but legibility. Compare the four endings on this page:
| Ending | Mechanism | What a holder could do |
|---|---|---|
| Aura | Dissolution vote authorising treasury redemption | Redeem through a front end kept alive for the purpose – but the underlying BAL stayed locked to April 2027 |
| Swapr | Token-shutdown vote with a published per-token rate | Claim a computable share of a fixed pool |
| DXdao | No dissolution vote; per-member redemption and burn | Exit individually, whenever – and 84% of supply eventually did |
| Across | Vote to transfer the protocol to a company, Across, Inc. | Sell $ACX for $USDC at a fixed price, or exchange it for company equity subject to identity checks and legal caps, until 8 January 2027 |
The third row is the interesting one, because it is the only mechanism on the list that did not require the organization to still be functioning. A dissolution vote is a collective act that presupposes a quorum, a working forum and someone to execute the result; the further a DAO decays, the less able it is to perform the ceremony that would end it cleanly. That is the argument for building the exit into the contract rather than into the constitution – the position developed at length under rage quit and exit rights, and the reason treasury management and dissolution are really the same design question asked at different times.
A caper has no dissolution procedure, because it has nothing to dissolve to. There is no wind-down among the five typed executive proposal kinds the contract recognises – PAYOUT, INVEST, DIVEST, METADATA and UPGRADE – and no function anywhere in the component that closes it, freezes it, or empties it. A caper cannot vote itself out of existence.
What it has instead is the redemption running continuously from genesis. Exit is a public method (exit() in contracts/logic/src/lib.rs) that any member who has cast a ranked ballot may call at any time: it takes their governance tokens and their soulbound vote tokens, computes their share as the same canonical weight that sized their vote, and pays it out of the treasury and the bonding-curve reserve. It requires no proposal, no quorum, no working forum and no counterparty – the three things a decayed DAO can no longer supply.
That inverts the sequencing problem this page describes. In a conventional DAO the exit is the last collective act, attempted at the moment the organization is least capable of collective action; in a caper it is a standing individual right that never depended on the organization being healthy. A caper does not end with a vote and a redemption window. It ends the way DXdao actually ended – one member at a time, whenever each decides – except that this is the designed path rather than the fallback after the designed one failed.