The Dank Bank Vaults

Verdict

Two protocols, not one

Architecture

Three layers, only one of them bespoke

What looks like one system is three, stacked. Understanding which layer a given risk or opportunity lives in is the whole game — the vault layer is immutable and permissionless, the market layer is upgradeable by one signature.

Roster

There are 21, not 14

An exhaustive Multicall3 sweep of all 4,616 vaults on Fractional's v1.1 factory plus all 2,503 on v1.0, filtered by curator, returns 28 Dank Bank vaults, of which 21 are live meme tokens. Dank Bank's own committed subgraph manifest lists exactly those 21, and data/tokens.json currently carries 15 of them. Seven are missing from the roster DeVamp carries — including the two cheapest complete buyouts in the entire system.

Reserve price is the full ETH cost to open a 3-day buyout auction for the whole NFT. “No quorum” means start() is uncallable at any price because fewer than 50% of the supply has voted a price — minVotePercentage is 500 ‰ and can never be lowered. Every vault is auctionState() == 0, vaultClosed() == false.

Asset register

All 21 vaults and what each one holds

Every row verified by token(), id() and ownerOf() on the vault. ownerOf returns the vault's own address for all 21 — every NFT is genuinely escrowed, and none has ever left. Ten are one-token NFT Basket wrappers, so the asset you ultimately own sits one layer deeper; the inner token is listed too.

Links go to OpenSea for the NFTs, Etherscan for the vault contracts and Polygonscan for the bridged children. TRIANGLE's Rarible token id is 78 digits (it encodes the creator address), truncated here — the full value is in the vault's id().

Mechanism

Uniswap v2 maths, fabricated reserves

Every pool lives in mappings on one contract, keyed by token address. There is no pair object, no factory, no getReserves(). The token reserve is just balanceOf(market). The ETH reserve is two numbers added together, and only one of them is real:

mapping(address => uint256) public virtualEthPoolSupply;   // fictitious, set once at initPool
mapping(address => uint256) public ethPoolSupply;          // real ETH actually held

function getTotalEthPoolSupply(address token) public view returns (uint256) {
    return virtualEthPoolSupply[token] + ethPoolSupply[token];
}

With FEE_MULTIPLIER = 311 and MULTIPLIER_SUB_ONE = 310, where T is pooled tokens and E is virtual + real ETH:

That is algebraically Uniswap v2 with γ = 310⁄311 — a 0.321543% fee taken on the input side, retained in the pool for LPs. There is no feeTo and no protocol cut. Note the contract's own comment claims ~0.42069%; it is wrong for the constant it ships.

Why no aggregator could ever route to it

You were right that routing friction killed it, and the reason is structural rather than a missing integration. Five independent blockers, all confirmed against source:

• Nothing to discover. One contract, mappings keyed by token. factory() and getReserves() both revert. No getPair, no allPairs, no per-pair address for an indexer to enumerate.
• Bespoke trade surface. buy(address,uint256) and sell(address,uint256,uint256) — no swapExactETHForTokens, no v2 swap(), no v3 exactInputSingle.
• Non-standard events. DankBankBuy / DankBankSell, never Swap / Sync / Mint / Burn.
• Native ETH only. No WETH pair, no ERC-20 quote asset.
• ERC-1155 LP, so generic position tooling cannot see the liquidity either.

And the one that would have broken an aggregator even if it had integrated: the quote function prices against reserves the contract does not hold, then the execution reverts. A quote-then-execute router would have failed constantly.

Confirmed empirically — 147 pair lookups on Ethereum across Uniswap v2/v3 (all four fee tiers) and SushiSwap against WETH and USDC, plus 285 on Polygon, all return address(0) with working controls. No token has ever appeared on CoinGecko, CoinMarketCap or DexScreener.

Activity

Full decoded log history of the proxy: 958 trades — 597 buys, 361 sells — from 203 unique traders, 109.83 ETH bought against 110.08 ETH sold. By year: 9 in 2021, 745 in 2022, 65 in 2023, 116 in 2024, 15 in 2025, 8 in 2026. First trade 2021-08-22, last 2026-06-15. It is dormant but not dead.

The trap

53 ETH of quoted liquidity, 1.4 ETH of real

This is the single most important fact for anyone putting money in. Across your fourteen pools there is 53.0957 ETH of virtual reserve against 1.4054 ETH of real ETH — and every wei of that real ETH came from buyers, not from liquidity providers.

The proof is in the LP accounting. initPool mints initVirtualEthSupply + msg.value shares, and virtualEthPoolSupply is write-once. For thirteen of the fourteen pools, lpTokenSupply equals virtualEthPoolSupply to the wei — which means msg.value was zero at genesis. Only TRADE ever had real ETH deposited: 0.5956 ETH. In the whole history of the contract there are 27 LiquidityAdded events and exactly one LiquidityRemoved, and that one returned 0 ETH.

How much you can actually sell before the revert, measured by bisecting each boundary and round-tripping it through the quote function:

Spend 1 ETH on TRIANGLE and you receive 2,292.25 tokens. You can then sell back at most 0.751 of them, for a total of 0.000212 ETH. Treat every purchase here as one-way.

Routes

Four ways to take supply, and what each is worth

Pricing

What the AMM route costs

Computed from live reserves with the exact source formula and spot-checked against the on-chain quote function.

ETH cost to take that fraction of the market's own holdings, ordered by cheapest first. “AMM holds” is that slice as a percentage of total token supply, so buying 90% of the TRIANGLE pool for 16.74 ETH leaves you with 14.0% of all TRIANGLE. Slippage on TRIANGLE runs +0.4% at 0.001 ETH, +5.7% at 0.1 ETH, 1.54× spot at 1 ETH and 6.40× at 10 ETH.

Risk

Three things a single key can still do

Supply cannot be minted, transfers cannot be frozen, there is no blacklist, no fee-on-transfer, no pause and no expired deadline anywhere. The token contracts have no owner() at all. What remains is worse than a typical admin key, because it reaches the asset rather than the paper.

Polygon

19 child tokens, and the bridge home is open

19 of the 21 have Polygon supply; only SMI and FEDORA were never bridged. Every mapping is a standard PoS ERC-20; the L1 escrow balance in the ERC20Predicate is the Polygon circulating supply, and child totalSupply() matches it to the wei for 18 of 19.

HIDEYOKIDS is the one mismatch: L1 escrow exceeds Polygon supply by exactly 0.1 token — one Polygon-side burn whose L1 exit was never claimed.

Where it sits, and how to get it back

For 18 of the 19, the Polygon float is ~95–100% held by the Polygon AMM plus two EOAs — and each token's Polygon holder is exactly that vault's L1 curator, 19 for 19. This is not a distributed retail float; it is one curator key per vault carried across the bridge. TRADE is the sole outlier: 0.14% in the AMM, 0.25% with the curator, and 99.58% genuinely distributed — chiefly contract 0x8cd79898… (38.99%), EOA 0xbdbae5d1… (25.95%) and EOA 0xccd8f8cc… (22.44%).

The exit works. Verified live: withdraw(uint256) on a child contract simulated successfully from a 600-TRIANGLE holder; an oversized amount reverted ERC20: burn amount exceeds balance; a checkpoint was 2.4 minutes old at measurement. Settlement is checkpoint latency (~30 min–3 h) plus a manual exit() proof on Ethereum. There is no 7-day challenge period — that applies to legacy Plasma exits. Polygon PoS is not deprecated; the July 2026 sunset was zkEVM Mainnet Beta, a different chain.

Execution

The transaction, exactly

The dead website is irrelevant. Every historical buy from 2022 to June 2026 is an EOA calling this selector directly — no router, no WETH wrap, no approval, no deadline, no signature, no allowlist.

Buy — mainnet

to:    0x0988Fb1C153a6f6492218Ead6b17fd795d6D18F9
value: <ETH to spend>
data:  0xcce7ec13 || abi.encode(address token, uint256 minTokensOut)
gas:   ~90,000        // observed gasUsed 86,847

Quote first — always

calculateBuyTokensOut(address token, uint256 ethIn)   0xa615e05a
calculateSellEthOut  (address token, uint256 tokensIn) 0x8108dd21   // lies — see §The trap
ethPoolSupply        (address token)                   0x24eb07f5   // the real number

Sell — two steps, and mostly futile

1. approve(0x0988Fb1C…, tokensIn)   // on the meme ERC-20; safeTransferFrom, so mandatory
2. to 0x0988Fb1C…, value 0
   data 0x6a272462 || abi.encode(address token, uint256 tokensIn, uint256 minEthOut)

Buyout auction — the route that gets you the NFT

reservePrice()  0xdb2e1eed   // full ETH cost; msg.value must be >= this
votingTokens()  0x9a4e6d34   // must be >= 50% of totalSupply() or start() reverts
start()         0xbe9a6555   payable
bid()           0x1998aeef   payable   // min increase 5% (minBidIncrease 50/1000)
end()           0xefbe1c1c             // after auctionEnd; transfers the NFT to `winning`
redeem()        0xbe040fb0             // burns 100% of supply, takes the NFT directly

Everything above was proven by read-only eth_call with a funded from override at block ~25,833,276. buy(token, 0) with 0.01 ETH returned success on TRIANGLE, ERMAGHERD, PANDA, MOMMY and TRADE; with 10 ETH on TRIANGLE, SUCCESS, CROISSANT and TRADE. sell(TRIANGLE, 100e18, 0) reverts with DankBankMarket: Market has insufficient liquidity for the trade. No transaction was broadcast and none will be — I can research and simulate this, but I can't execute trades or transfers on your behalf, and nothing here is investment advice.

Gaps

What I could not settle

1. Who holds the two curator keys 0xd4e3a3cb and 0x30653411, and the two signers on the 1-of-2 upgrade Safe. Between them they can collapse the reserve price on eight vaults, withdraw all pooled ETH, and replace the AMM. This is the highest-value follow-up before committing capital — it is the difference between a dormant project and an active one.
2. Who holds the 19 Polygon child tokens. Nobody has run a holder query against them. For TRADE, at 97.74% bridged, this is the acquisition question.
3. What drained the Polygon AMM's USDC.e to $0.002 against an 8.99M book. The leading untested hypothesis is 18-decimal seed figures against a 6-decimal payment token.
4. A possible Polygon staking contract. The live frontend references dank-bank-staking-matic and -mumbai subgraphs. There is no staking repo in the org and no address in any config. It may hold user funds. Nobody has looked.
5. 4,206,900 FEDORA stranded in a failed MISO auction. finalizeTimeExpired() returns true, so a finalize() should release them — but who is permitted to call it is unverified.
6. Whether Safe 0x8569fCA4 (feistydao.eth, 5-of-9) is Dank Bank's treasury or FeistyDAO's. Its holdings — 13.4bn NFD, 100m DOG, ten Uniswap v3 positions — point to FeistyDAO as an outside participant, not the issuer. Unresolved; its 161 transactions were never decoded.
7. The five superseded AMM implementations were never disassembled. The 745 trades of 2022 — the bulk of all volume ever — cleared under a version nobody read.