The claim
Most backed tokens are backed by a sentence.
Somebody says there are treasuries. Somebody says there is a vault. What you can actually check is a promise on a website and a wallet you are told to trust.
ASSAY makes a narrower claim and makes it checkable. The vault holds PAXG — Paxos Gold, one fine troy ounce of London Good Delivery bullion per token, allocated to serial-numbered bars, attested monthly, redeemable with Paxos for metal. You can read the vault's balance yourself, in any block explorer, without asking anyone.
The division of trust is worth stating exactly, because it is the whole point. Paxos custodies the gold. That is their undertaking, supervised by the New York Department of Financial Services, and no line of Solidity strengthens or weakens it.
This protocol proves the rest: how much PAXG sits in the vault, that only a redeemer can remove it, that redemption cannot dilute anyone, and that the gold behind one token has never fallen. Those are arithmetic, and arithmetic does not require a press release.
The mechanism
Four moving parts. None of them can take your gold.
TouchThe hook takes a share of every swap, in ETH, on every shape of swap. It holds that ETH and can send it to exactly one place.
MeltAnyone may call the refinery. It draws a bounded batch, swaps it for PAXG on the Uniswap v4 gold pool, and refuses any fill more than 3% below the pool's own price — a floor the caller can raise but never lower.
DeliverThe gold lands in the vault, credited by balance delta rather than by a number anyone passed in. Fineness rises. The swap rate falls.
RedeemBurn your tokens, take your share of the gold. The units are destroyed, so the gold behind everyone else's tokens is untouched.
The refinery is deliberately not on your trade path. Converting inside every swap would make every trader pay for a second swap forever, would break when the gold book is thin, and would revert your trade for reasons that have nothing to do with you. Measured on mainnet: the ETH/PAXG pool holds about 122 ETH in range, so one ETH converts at 0.7% slippage and five at 3.3%. A fee-sized batch is comfortable; an arbitrary one is not.
The instrument
Redeeming cannot dilute anyone. Try to break it.
This runs the contract's real integer arithmetic — same operations, same rounding. Trade, refine, then redeem as hard as you like and watch the fineness.
Gold held0.000000troy oz PAXG
Fineness0.000e+0troy oz per ASSAY
Supply1,000,000,000nothing burned
ETH waiting0.0000held for the refinery
A simulation of the deployed arithmetic. No wallet is involved and nothing here touches a network. The gold price used for melting is the one measured on mainnet, 0.5610 PAXG per ETH, with the contract's 3% floor applied.
The rate
The more gold it has bought, the less it charges.
The swap rate opens at 3.00% and falls toward 0.50% as gold accumulates. The input is gold ever delivered, not gold currently held — so redeeming cannot push the rate back up, and the rate is monotonically non-increasing for the life of the protocol.
That distinction is the difference between a mechanism and an exploit. Pricing off the vault's current balance would let a large redemption raise everyone's fee. Pricing off the pool's spot price would let a flash loan move it inside one transaction. Lifetime delivery moves in exactly one direction, by actually buying gold.
Gold delivered0.00troy oz, lifetime
Rate3.0000%of the ETH leg
Cost to trade3.30%including the 0.30% LP fee
What is provable
The honest boundary.
Certificate of fineness — scope of proof
What this protocol proves, and what it does not
ON CHAIN
The vault's gold balance. A PAXG balance any explorer will show you, at any block.
ON CHAIN
That only a redeemer can remove it. Fifteen functions, none of which move gold out except
redeem, and no owner exists to add one.ON CHAIN
That redemption cannot dilute. Fork-measured across a mixed run of trades, refines and redemptions: fineness rose 12 times and fell 0. A single redemption moved it 0 wei.
ON CHAIN
That the rate can only fall. It is a decreasing function of gold ever delivered, and nothing decreases that number.
PAXOS
That the gold exists in a vault. Paxos Trust Company's undertaking — allocated bars, monthly attestation, NYDFS supervision. Verify it at Paxos, not here.
PAXOS
That PAXG stays redeemable for metal. Also Paxos's, and it is the load-bearing external dependency of this entire design.
NEITHER
That the token's market price goes up. Nothing here is a claim about price. The vault is a floor, not a forecast.
If Paxos fails, the vault holds a token that no longer represents gold. That risk is not hedged, not insured and not diversified, and it is stated here rather than buried in a footnote because it is the single largest thing that can go wrong.
Absent, deliberately
What was never fitted.
no owner
no mint
no pause
no upgrade
no proxy
no fee setter
no withdrawal
no blacklist
no max wallet
no cooldown
no transfer tax
no reflection
no rebase
no keeper
no oracle
no admin key
no rate that rises
no path to the gold
no privileged reader
no re-armable gate
The hook's six liquidity permission bits are clear, which means the PoolManager never consults it when anyone adds or removes liquidity. It cannot block, tax or delay an exit — and that is a property of the hook's address, fixed by CREATE2, not a promise in a comment. You can check it yourself: take the address, mask the low 14 bits, and read them.
What it costs you
Stated plainly.
Fineness starts at zero
Nothing is backed at launch. The vault fills only from trading, so on day one you are buying a mechanism, not a reserve. Buying at market and redeeming immediately loses money by design — otherwise the vault would be a faucet.
It costs 3.30% to trade at open
3.00% to the hook plus the pool's 0.30%. It falls toward 0.80% as gold accumulates, but early traders pay the most and that is the honest order of things.
The gold book is thin
About 122 ETH in range on the v4 ETH/PAXG pool. Batches are capped at 1 ETH for that reason. If gold liquidity dries up, refining stalls — the ETH simply waits, and nothing is lost.
Paxos is a single point of failure
One custodian, one jurisdiction, one token. PAXG sits behind an upgradeable proxy. The vault is written to survive a transfer fee being switched on, but not Paxos ceasing to honour redemption.
The record
Measured, not asserted.
Every number on this page comes from a simulation run against a fork of Ethereum mainnet, using the real Uniswap v4 PoolManager and the real Paxos Gold pool — not a mock.
ETH converted to gold on the live pool → 0.055141 PAXG at 0.5567/ETH, against a 0.5610 spot
Redemption drift → 0 wei
Mixed run of 12 trades, refines and redemptions → 12 raised fineness, 0 lowered it
All four swap shapes charged → exact-in buy, exact-in sell, exact-out buy, exact-out sell
Gas on a warm buy → 131,049 · worst cold buy 216,581
Vault functions that can move gold out → 0 of 15
Gas drift as the vault fills → +92 across a 3.7× increase in delivered gold
Reproduce it with node sim_assay.js against a Cancun mainnet fork. If a number here disagrees with that run, the run is right.
The preparation
Six contracts.
AssayToken$ASSAY1B struck once, no mint, public burn
Vaultthe reserveholds PAXG, redeem is the only exit
AssayHookflags 0x20CCtakes ETH on all four swap shapes
Refinerypermissionlessmelts ETH to gold, price-floored
CrucibleCREATE2casts the hook or reverts
Assayerread-onlythe whole certificate in one call
Reserve assetPAXG0x45804880De22913dAFE09f4980848ECE6EcbAf78
PoolETH / ASSAY0.30% fee, tick spacing 60 — the Uniswap app's own default, so no URL editing
Opening rate3.00%hard ceiling, checked at construction
Floor rate0.50%where a well-stocked vault lands
Halfway at10 troy ozgold delivered, lifetime
Refine batch1 ETH max3% max slippage, 0.50% caller bounty
Addresses—not deployed; any address circulating is not ours
Compiler0.8.28viaIR, runs 200, cancun