Liquidity vaults
What a concentrated range actually costs when the stock falls, why the position becomes all stock below its lower bound, what the lock means, and why this liquidity is temporary by design.
Spec v0.9.1, reviewed 2026-09-08
A liquidity vault takes your stock and your fyUSD, puts both into one Uniswap position, and lets that position buy the stock as it falls and sell it as it rises. It pays you in staked FBR. It is the most exposed position in the protocol, and the numbers below are the reason.
The mechanism is on Liquidity vaults and the user path on Provide liquidity.
You are a Stability Pool depositor without the discount
This is the honest framing, and the design record uses it. A pool depositor buys collateral at the liquidation price minus a discount of 2% to 4%, only when a position is actually under water, and can withdraw at any time. A vault provider buys the same stock at the market price, continuously, all the way down, with a lock, and the trading fee goes to the protocol rather than to them.
Risk
For the same capital and roughly the same role, the vault has worse economics than the Stability Pool. What it has that the pool does not is a quoted price on the way down, and a return path from the stock back to fyUSD that did not exist before. Those are protocol benefits, and the provider is paid for them in a token whose price is not fixed.
What a fall actually costs, tier 1
A tier 1 vault holds a range from 40% below the centre to 25% above it, and starts mostly in fyUSD. These figures are the position's value against the two obvious alternatives.
| The stock falls | fyUSD left in the position | Value against the deposit | Against holding both | Against holding only fyUSD |
|---|---|---|---|---|
| 10% | 77% | −4.0% | −0.8% | −4.0% |
| 20% | 53% | −9.7% | −3.6% | −9.7% |
| 30% | 28% | −17.6% | −8.9% | −17.6% |
| 40%, the lower bound | None | −28.1% | −17.6% | −28.1% |
| 50% | None | −40.1% | −28.7% | −40.1% |
| 80% | None | −76.0% | −67.8% | −76.0% |
Tier 2, with a range of 55% and 35%, on the same basis: a 20% fall gives −8.3% against the deposit and −2.5% against holding both, with 68% of the fyUSD still there; 30% gives −14.6% and −6.3%, with 50% left; 50% gives −33.2% and −21.5%, with 11%; 55%, the lower bound, gives −39.5% and −27.7%, with none left; 80% gives −73.1% and −64.7%.
Tier 3, at 65% and 45%: a 20% fall gives −7.8% and −2.0%, with 74% of the fyUSD left; 30% gives −13.4% and −5.1%, with 60%; 50% gives −29.5% and −17.4%, with 28%; 80% gives −70.2% and −61.1%.
For comparison, a full-width position loses 0.6%, 5.7% and 25.5% against holding both at falls of 20, 50 and 80 per cent.
Risk
Concentrating the range multiplies the loss against holding both by roughly ten, inside the range. That is not a defect: it is what buys the depth. It is also the whole of what you are being paid for.
Below the lower bound you hold nothing but the stock
Inside the range the position is a mixture. Below 40% on tier 1 it is one hundred per cent stock: every unit of fyUSD has been spent buying the fall. Above 25% it is one hundred per cent fyUSD: the stock has all been sold on the way up.
Out of range, the position holds no liquidity at the trading price. The pool stops quoting, the vault's contribution to measured depth goes to zero, and its price stops being read as a source. It is absent rather than wrong, which is the correct behaviour, and it means the vault has stopped doing the job it was paid for until someone recentres it.
Recentring is permissionless and needs the price to be at least 15% away from the centre, with at least 24 hours since the last one, in the full price regime. Nobody is obliged to call it.
Both a deposit and a recentring are refused outright while the vault's own pool sits more than 1% from the composite price. That is a protection rather than an outage: entering a pool that is briefly off-price means being arbitraged in the next block, at your expense. Waiting a few minutes is the whole remedy.
A deposit no longer has to match the vault's ratio. What can be paired is taken, and the excess of whichever asset you brought too much of is handed straight back in the same transaction.
Risk
A stock that triples in a year is recentred repeatedly, and each recentring sells more of it. Over six months of a stock going up three times, the design's own worked case leaves the vault at about +45% against +200% for holding the stock. Selling into a rise is what a range position does, and the buy wall it leaves behind is exactly what a liquidation needs. It is still a large amount of upside that the provider does not receive.
The lock does not lift
Four steps, from 7 days to 365 days, carrying a reward multiplier up to 3.0×. Locked shares cannot be transferred and cannot be withdrawn before their date.
There is no early exit, at any price. No penalty schedule, no buy-back, no exception. The single mechanical release is when the branch shuts down or the whole protocol enters its terminal mode, at which point every lock falls at once and anybody may trigger it.
Guarantee
Withdrawing unlocked shares is never blocked: not by a price regime, not by a freeze, not by a dead source, not by the Closer key, which does not know the vaults exist. What comes back is your fraction of what the vault holds, in kind, whatever that mixture happens to be.
Rules R-19.15.3, invariant 61
The reward is staked, and the price is not fixed
The vault class of the season budget is delivered as staked FBR and never as liquid FBR. From there it leaves by the exit grid: nothing after 7 days, or 70% withheld if you want it immediately.
Two consequences. Your reward carries the FBR risks in full, which are on FBR, including seizure while it is staked. And the reward you are quoted in units is worth whatever FBR is worth on the day you sell it, which nobody sets and nobody promises.
A quarter of the trading fees does come back to locked providers, in fyUSD, independent of the FBR price. On the modelling in the design record that is in the region of half a per cent to one and a half per cent of the position per year.
This liquidity is temporary, and the protocol says so
The programme is paid for out of a season budget that halves roughly every two and a half years, and steps down sharply after the fourth season. What it buys, per active branch in the first year, is a floor of a few tens of thousands of dollars per one per cent of price move. What the branch ceilings assume is between two hundred and fifty and three hundred and seventy-five thousand. Widening the ranges made that gap larger rather than smaller, and it was the right trade: a vault that still has liquidity at the trading price is worth more than one that is deeper at a centre the price has left.
Risk
The programme cannot buy the depth the risk parameters are sized against. It buys a floor, and the documentation is required never to present it as the protocol's depth. The honest expectation in the design record is that vault deposits fall by half to four fifths between the fourth and the eighth season, staggered by the locks, as the emission decays.
Rules R-19.15.9, R-15.4.1 (15)
Nothing in the protocol depends on that not happening. Measured depth sums every pool, the vault pool is disqualified as a price source until it is genuinely deep, no ceiling is reserved for it, and an empty vault makes every one of its functions do nothing.
A vault's pool counts toward its branch's borrowing ceiling only for the share of it that is locked, meaning lots more than a day from their unlock date. Depth that can leave tomorrow does not raise what the branch is allowed to lend today, and an interface that publishes one depth figure without that split is overstating the branch's capacity rather than the vault's size.
Three more things that can go wrong
The stock stops transferring. The vault holds stock and fyUSD in one position, so an issuer freeze reaches both. A withdrawal still returns the fyUSD side; the stock side stays claimable separately and is paid out when the token moves again; and after 7 days of that state every lock in the vault falls, so a provider on a 365 days lock is not held by the lock. The frozen stock stays frozen, which is the part no mechanism can reach. This is on Issuer and freeze.
Uniswap takes a cut. The pool contract's owner can enable a protocol fee of up to a tenth of a per cent per swap, taken before the liquidity fee. At the worst setting that is a fifth of the vault's fee revenue, and neither Fyber nor the provider can prevent it.
The width was bought with depth. The ranges are measured rather than guessed: five years of history across twenty names, over which a vault was never left without liquidity at the trading price and still held 26% of its fyUSD at the worst episode. That is the case for them. The price is concentration, and therefore depth at the centre: 19 to 30% less than the narrower ranges the earlier design carried. On SPY that is 7 to 15 thousand dollars per one per cent in the first year, where the narrower range would have bought 10 to 20.
They are frozen at deployment like everything else, so a name whose behaviour departs from its five-year history keeps the range it was given.
Peg and the swap module
What holds fyUSD near a dollar, what happens above and below, the cap that never rises, what a USDG depeg does, and the loop between the liquidity vaults and the reserve.
FBR
A token with no claim on anything, that can be seized to cover somebody else's bad debt, that costs to leave, and that can be worth nothing without a borrower or a depositor noticing.