An ERC-20 whose contract holds the basket itself. One U is a fixed list of quantities; mint delivers that list into the vault, burn hands it back. No rate oracle stands between you and redemption — and the two honest complications Ethereum forces, gas and permissioned legs, are stated on page one rather than buried in an appendix.
Companion to the design spec and methodology · basket Revision R3, 2026-09-20 · sister deployment on Solana: glu.booga.me
Every leg of Revision R3 — gold, euros, equities, property and bitcoin — is a live token on Ethereum today. Nothing in the basket is waitlisted, synthetic, or standing in for something that has not shipped. That is the case for building here. The other half of the truth: Ethereum is the expensive chain, and two of the six legs come with allowlists attached.
| What the basket needs | What Ethereum has |
|---|---|
| Euro stablecoins · 29% | EURC · EURS · agEUR — the one non-dollar currency with real tokenized depth; held as a dollar hedge, not for its own return |
| Gold · 35% | PAXG · XAUt — redeemable claims on allocated bars, with real DEX depth on both |
| US equities · 19% | bCSPX · Ondo Global Markets — real and tradable, thinner than the rest, and transfer-restricted in places |
| Bitcoin · 7% | WBTC · cbBTC · tBTC — three independently custodied wrappers, all deep; the small weight is a risk decision, not a supply one |
| Real estate & private credit · 10% | RealT · Centrifuge — the thin end of the shelf. Capped at 10% for exactly that reason, and the cap is in the contract |
| Price feeds | Live and plentiful — but not needed to mint or burn. The vault holds the assets, so feeds serve display and the charter's risk checks only |
| Cost of a mint | The honest problem. A six-leg atomic mint on L1 is expensive. Netting, a minimum mint size and an L2 deployment are the answers, and none of them is free |
| Verdict | Nothing is missing from the shelf. The remaining work is execution and permissions — gas, slippage and allowlists — not waiting for assets to appear |
One U is a fixed list of quantities, published at a revision — the same idea as an SDR or an ETF creation unit. Mint by delivering that list in kind, or by delivering USDC which the contract atomically swaps into the exact quantities on Ethereum DEXes. Burn to get the list back, in kind or swapped out to USDC. Because the vault holds the assets themselves, the price of U is simply what its contents cost right now — Σ qᵢpᵢ — and no rate oracle is needed to mint or burn. That removes the entire class of failure where a feed goes stale or wrong and redemption stops with it.
1 U := a fixed quantity of each leg below — delivered into the vault at mint, handed back at burn. Price of U = Σ qᵢpᵢ. Per-leg invariant: vaultᵢ ≥ qᵢ × S
The invariant is stated per leg, not in aggregate: holdings of each asset must be at least the quantity-per-U times the supply S. That is strictly stronger than a value invariant — a vault cannot quietly drift into being long one leg and short another while the total still looks right. It is checked at the end of every mint and every burn. Full list and rationale in the mechanism §1.
| Leg | Tokens held | R3 weight |
|---|---|---|
| GOLD | PAXG · XAUt | 35% |
| EUR | EURC · EURS · agEUR | 29% |
| SPX | bCSPX · Ondo Global Markets | 19% |
| HOUSE | RealT · Centrifuge | 10% |
| BTC | WBTC · cbBTC · tBTC | 7% |
Measured in purchasing power rather than dollars, R3 backtests to +5.86%/yr real over 20 years — 3.13x — with 7.9%/yr volatility and a worst drawdown of 18.6%. Over the last ten years: +8.80%/yr real, 2.32x, volatility 7.8%, worst drawdown 13.4%. Those drawdowns are the point of printing them. U goes down. It has gone below entry for holders who bought at the wrong moment, and the charter's job is to bound how far, not to pretend otherwise.
Dollar stablecoins were removed from the universe outright at R2. USDC and its peers are the deepest on-chain money there is, and that was never the question: idle dollar cash lost 1.71%/yr of purchasing power over the window while earning nothing, and the 10% floor that forced the vault to hold it was the most expensive line in the old charter. Redemptions are now paid in kind, or out of the T-bill leg, which is the vault's near-cash and pays to sit there. Dollar exposure did not leave with it — tokenized T-bills are dollar instruments, so the basket remains about 37% dollar-denominated. What left was the part that only melted.
ETH (stETH · wstETH) remains eligible and remains at zero: +4.65%/yr real, marginally better than gold, at 58.6% volatility and a 90.3% drawdown — four times the risk for no additional return. Its ceiling is already written into the contract, so a future revision can admit it without a code change if the evidence turns. Euro stablecoins went the other way, and the reversal is worth reading: under R1's dollar floor the optimizer declined them, because euro cash lost 2.11%/yr. Without the floor it buys 12% of them — not for the return, which is still negative, but because euros move against the dollar, and that hedge buys the volatility room to carry a heavier bitcoin leg.
The rule from the methodology applies here at maximum strength: print the cost, don't bury the leg. A basket that names its drawdowns and its allowlisted holdings is honest; one that quotes only the compound return is not.
A standard ERC-20 plus a non-custodial vault module built on audited components. It holds the six legs, mints against the quantity list, burns back to it, and enforces the per-leg invariant, the ceilings and the USD floor in-contract. Upgradeable behind a timelock — except redemption, which is not upgradeable at all.
USDC mints route through 1inch / CoW Swap with a slippage band per leg. Every leg fills inside its band or the whole mint reverts; partial baskets are never minted. Caps are sized to live pool depth and re-sized as depth moves.
A web app over WalletConnect covers any Ethereum wallet. L1 is the settlement venue for the vault; a Base/Arbitrum deployment carries cheap retail mints, netting them and settling to the L1 vault, because paying L1 gas on a small mint is indefensible.
In kind — deliver all six legs in the exact quantities, the route market makers will use — or in USDC, and let the contract buy them. Both end in the same state; the second one costs more, and the quote says how much.
USDC is routed into the exact quantities in one transaction, each leg inside its slippage band. A minimum mint size applies: below it, gas dominates the trade and the user simply loses money. Batching and netting across pending mints reduce the per-user cost.
U is issued only once vaultᵢ ≥ qᵢ × S holds for every leg at the new supply. Mint fee f; terms published separately. Gas and realised slippage are printed in the quote before signing, not discovered afterwards.
Burn U and take the list back in kind, or have it swapped out to USDC at the same per-leg bands. No pause switch exists for redemption, by design — not for stale feeds, not for governance, not for the advisor. The invariant is re-checked after the burn.
Redeeming in kind costs gas and the fee f. Redeeming to USDC adds the slippage of selling six legs, which is highest on the thin ones — property and credit first, equities second. Taking the assets out yourself is always the cheaper door, and it is always open.
An AI advisor proposes the next quantity list — yearly, or sooner when a newly tokenized asset qualifies — maximizing backtested purchasing-power growth subject to a charter the contract enforces rather than merely states: volatility at or under 6%/yr, worst drawdown at or under 12%, the per-leg ceilings below, turnover at or under 20% of the vault per revision, and at most one revision per quarter. Every proposal publishes its full reasoning, then sits in a public 7-day timelock before it can take effect, so anyone who disagrees can read the argument and leave before it binds. The advisor can never touch redemption rights, fees, or the invariant.
The ceilings below are not policy statements — the contract rejects any quantity list that breaches them, including one the advisor proposes. They follow liquidity, not conviction: how much of a leg the vault could actually sell into a bad week.
| Leg | Liquidity tier | Ceiling | R1 |
|---|---|---|---|
| Gold | deep | 35% | 35% |
| EUR stables | mid | 40% | 29% |
| US equities | mid | 25% | 19% |
| Bitcoin | deep | 15% | 7% |
| ETH | deep | 15% | 0% |
| Real estate & credit | thin | 10% | 10% |
| Cash & government debt | Dollar stablecoins removed at R2, US Treasury bills at R3. The unit rests on no sovereign’s balance sheet — and gave up its only drawdown control to do it. Redemptions are paid in kind or from the euro leg | — | |
| Risk | Guardrail |
|---|---|
| Gas on L1 | The defining cost of building here. A six-leg atomic mint is expensive: mitigated by batching and netting pending mints, a minimum mint size, and an L2 deployment (Base or Arbitrum) that settles to the L1 vault. It is printed in the quote, never absorbed silently |
| Slippage on thin legs | Per-leg slippage bands with the whole mint reverting if any leg misses; mint caps sized to live pool depth. Property and credit are the thin end and carry the 10% ceiling for it; equities are capped at 25% |
| Permissioned legs | The honest constraint: BUIDL and several tokenized-equity and RWA tokens carry transfer restrictions or KYC'd allowlists, so a permissionless vault cannot simply hold them. Retail-accessible instruments of the OUSG / USDY kind are preferred wherever they cover the same exposure. Where they do not, the leg needs a qualified-custody or permissioned-wrapper path, and that path is a real dependency on a named third party — disclosed as such. Until it is proven in production, that portion of the leg is not counted as live backing |
| Stablecoin depeg | The USD leg is 10%, spread across USDC, USDT and PYUSD rather than one issuer. A depeg passes through to the price of U honestly instead of being papered over |
| Oracle failure | Structurally contained: feeds are not in the mint or burn path at all, because the vault holds assets rather than a claim on a rate. They drive display and the charter's risk checks. A stale or wrong feed can delay a revision; it cannot delay a redemption |
| Advisor overreach | The charter is enforced in-contract — volatility ≤ 6%/yr, drawdown ≤ 12%, ceilings above, turnover ≤ 20% per revision, ≤ 1 revision per quarter — behind a public 7-day timelock with published reasoning. Redemption rights, fees and the invariant are outside its reach entirely |
| Contract bug | Independent audits before the mainnet pilot, a supply cap and per-transaction cap during it, timelocked upgradeability for everything except redemption, and a vault that is non-custodial throughout |
| Legal | Same reality as the design spec §7: a basket-redeeming token is not a payment stablecoin; testnet first, capped pilot, geo-notice, no yield paid |
| Stage | Ship | Proves |
|---|---|---|
| E-1 | Testnet deployment. The ERC-20 and its vault, six mock legs, mint and burn both in kind and via swap, and the public timelock contract that future revisions must pass through | The whole loop end to end: quantity list → deposit → invariant check → burn back, with no oracle anywhere in the path |
| E-2 | Mainnet pilot, small cap. A real vault holding real legs, the permissioned-leg custody path proven in production rather than assumed, per-leg slippage bands live, and caps sized to observed pool depth | That the allowlisted legs can actually be held and redeemed, and what a mint really costs once gas and slippage are counted |
| E-3 | First live revision. The advisor's proposal published with its reasoning and executed through the 7-day timelock, alongside the L2 deployment that makes small retail mints affordable | That the basket can change under its charter without anyone's exit being touched — and that the cheap door exists |