fxSAVE — Risk Report

Elevated risk · 4.5/10

Category: Vault share | Underlying: fxSP (the f(x) Stability Pool), which redeems into fxUSD | Issuer: f(x) Protocol (AladdinDAO)

Summary

fxSAVE is an ERC-4626 vault share issued by f(x) Protocol on Ethereum. It is widely described as “saved fxUSD”, and that description is wrong in a way that matters: the vault’s asset() is not fxUSD but fxSP, the f(x) Stability Pool token. The Stability Pool is the protocol’s liquidation backstop — it absorbs collateral from liquidated positions in exchange for burning fxUSD, and pays depositors for taking that role. fxSAVE is therefore a yield-bearing claim on a loss-absorbing pool, and its worst day arrives on the same day as the worst day for the collateral behind fxUSD.

The token has no peg. One fxSAVE was worth 1.108142 fxSP on 2026-09-22 and the exchange rate rises as the pool accrues. At a quoted price of $1.1189 and 69,462,472 shares outstanding, the market capitalisation is about $77.7M. Redemption is immediate, permissionless and uncapped, and the accounting reconciles exactly — but it pays out in fxUSD, whose own protocol redemption is refused and whose market can yield at most about $4.65M of USDC at any size, and an instant redemption realises about 4.3% less than the quoted price.

Roughly 27% of all fxSAVE — about $21.0M — sits in Morpho Blue as collateral for USDC loans at 86% and 91.5% loan-to-value. The 4.5/10 overall score reflects a well-constructed wrapper around a pool designed to lose value under stress, whose exit is only as good as the pool it lands in.

1 · Stability — 5.0

fxSAVE is a NAV token rather than a pegged one. Each share represents a growing claim on fxSP: convertToAssets of one share returned 1.108142 fxSP on 2026-09-22, and the rate rises as Stability Pool rewards accrue. Deviation from one dollar is the wrong lens for it; the relevant questions are whether the NAV moves monotonically and what the NAV is a claim on. On the first, the mechanism is functioning as designed and the rate has only risen.

On the second, fxSP is not a stablecoin. The Stability Pool exists to absorb collateral from liquidated f(x) positions, burning fxUSD to do so. In calm conditions that is a source of yield; in stress it is a conversion of the depositor’s dollar-denominated claim into the volatile collateral that was just liquidated. The value a fxSAVE holder carries is a claim on a buffer that is drawn down precisely when the protocol’s collateral is under pressure.

Risks and limitations. fxSP already redeems below par: previewRedeem of one fxSP returned 0.976407 fxUSD on 2026-09-22, about 2.4% under, and the instant path returns 0.966643. Whether that discount is a base fee or accumulated liquidation losses has not been established, and the two readings imply materially different things about how much the pool has already absorbed. That question is open.

2 · Backing — 4.5

fxSAVE’s assets are 76,974,259.94 fxSP against 69,462,472.31 shares (2026-09-22). fxSP in turn is a claim on the Stability Pool’s holdings, which convert into fxUSD on redemption, and fxUSD is over-collateralised by wstETH and WBTC held against leveraged positions in f(x)‘s CDP markets. So the backing passes through two layers before it reaches any external asset, and the external asset is volatile crypto collateral rather than a reserve.

The bookkeeping between the layers is exact. fxSAVE holds no fxSP directly; its assets sit in a segregated custody contract whose balance of fxSP gauge tokens matched totalAssets() to the wei on 2026-09-22, and the gauge held 99.68% of all fxSP (78,721,590.46 of 78,971,349.16). Nothing is unaccounted for, and confirming it requires only balance reads.

Risks and limitations. The collateral behind fxUSD is pledged against leveraged positions rather than held idle, so the backing’s worst day and fxSAVE’s worst day are the same event. The Stability Pool converts into that collateral by design during imbalance — that is the risk a depositor is paid to carry, not a defect, but it means fxSAVE cannot be read as a claim on stablecoins. The 2.4% sub-par redemption on fxSP is unseparated between fee and absorbed loss, and that is the reason the score sits half a notch below fxUSD’s own backing.

3 · Liquidity & Exit — 3.5

A holder can leave two ways. The spot route is thin: a single Curve pool, fxSAVE/scrvUSD at 0xb6E4821c6fCABe32f5F452dfD3Ef20Ce2A3a48E2, held $535,283 of TVL on 2026-09-22 — 0.69% of market capitalisation — and three other listed pairs (DOLA, litUSD, sUSDaf) held effectively nothing. Those figures are pool TVL from Curve’s stable-ng factory listing, not depth at a slippage bound.

The redemption route is strong, and it was simulated rather than assumed. On 2026-09-22, redeem of 1,000,000 fxSAVE from a real holder returned 1,108,141.67 fxSP and 10,000,000 returned 11,081,416.68 — linear, uncapped, with the same call from the burn address reverting on ERC4626ExceededMaxRedeem as it should. maxRedeem for Morpho Blue’s entire 18,751,881.70 balance returned the full amount. From fxSP, instantRedeem into fxUSD returned 966,643.36 on 1,000,000 and 9,666,433.62 on 10,000,000, again linear; a second, delayed path (requestRedeem) also succeeds.

The binding leg is neither of those. It is the last hop, fxUSD to USDC. fxUSD’s own protocol redemption is refused (see the fxUSD report), so that hop is DEX-only, and it was measured end to end on 2026-09-22: a routed sell of fxUSD crosses 2% at between $4.25M and $4.30M, and proceeds peak at $4,648,576 USDC — beyond that size the router sells fxUSD for ETH and total USDC falls. About $4.65M is the most USDC obtainable at any size, a ceiling rather than a cost. Thirty-three Curve pools were enumerated and no Uniswap, SushiSwap or Balancer venue for fxUSD exists. Redemption converts a fxSAVE position into a fxUSD position, and that ceiling is what a seller then meets.

Risks and limitations. fxSAVE’s exit is worse than fxUSD’s own, despite ending in the same market, because of an overhang fxUSD does not carry: 90.55% of fxUSD sits locked in the Stability Pool and only about 1.87M is free float, and a fxSAVE redemption converts locked fxUSD into float that then meets the $4.65M ceiling. Morpho Blue holds 18,751,881.70 fxSAVE (27.0% of supply), and a liquidation of that book is a forced, simultaneous, one-directional flow of roughly 20.1M fxUSD against a ceiling a quarter that size. A wrapper whose holders are concentrated in one lending market exits worse than the asset it redeems into. The fxSAVE spot venues were read from Curve’s stable-ng factory only; the ladder is a point-in-time router measurement and moves with pool balances. Only Curve’s stable-ng factory was enumerated; Uniswap and Balancer were not, so the $13.0M venue total is a floor. And the instant path costs about 1.0% more than the patient one.

4 · Dependencies — 4.5

Five contracts sit between a fxSAVE holder and a stablecoin, and the ERC-4626 interface shows only the first:

Layer	Contract	What it holds (2026-09-22)
Share	fxSAVE 0x7743e50F534a7f9F1791DdE7dCD89F7783Eefc39	no tokens directly
Custody	vault 0x4bFc5292DA18e32ED2354350e07e417cE4796f9C (EIP-1167 minimal proxy, 45 bytes)	76,974,259.94 fxSP-gauge tokens
Gauge	fxSP-Gauge 0xEd92dDe3214c24Ae04F5f96927E3bE8f8DbC3289	78,721,590.46 fxSP (99.68% of supply)
Pool	fxSP 0x65C9A641afCEB9C0E6034e558A319488FA0FA3be	the Stability Pool
Stablecoin	fxUSD 0x085780639CC2cACd35E474e71f4d000e2405d8f6	reached via instantRedeem

fxUSD is the terminal dependency: it is what the Stability Pool pays out, what the only exit pool is quoted against, and what the collateral ultimately backs. Everything above it is f(x)‘s own plumbing.

Risks and limitations. fxSAVE and fxUSD share one ProxyAdmin and one timelock (see Contract & Admin), so a holder who treats them as separate legs is counting diversification that does not exist — a single control failure reaches both. fxSP itself has no independent market: 99.68% of it sits in one gauge, and most of that gauge is fxSAVE’s own custody contract, so its value is only ever realised through the redemption path described above.

5 · Contract & Admin — 4.0

The fxSAVE share is an upgradeable proxy. Its EIP-1967 implementation slot resolves to 0xe4031e271809d20074e4bef1caeefec5f710e8a6 and its admin slot to a ProxyAdmin at 0x9b54b7703551d9d0ced177a78367560a8b2edda4, whose owner() is an OpenZeppelin TimelockController at 0x68863fb8855b04509a835082478D6E3D0bE4E61a with a minimum delay of 259,200 seconds — 72 hours. Every address here was read from a call that returned it, and the timelock’s role table was read off the contract: proposer and canceller authority is held solely by a Gnosis Safe v1.3.0 at 0x26b2ec4E02ebe2F54583af25b647b1D619e67BbF requiring 6 of 9 signatures, all nine owners externally-owned accounts; the timelock is its own only administrator — the Safe and one deployer address held that role from deployment on 2026-04-06 until both were revoked on 2026-04-20 (block 24,920,489, transaction 0x8449887472c310415b1b17f5daff8ab72511a76616cb257ccc02605f48e027af), in the same action that granted the executor role to the zero address. That action tightened the timelock’s own administration and nothing else. Shortening the delay is itself subject to it — updateDelay succeeds only when called by the timelock, so a reduction must be scheduled and wait the full 72 hours in public.

The delay covers code upgrades, and that is the extent of it. The fxSAVE contract’s own DEFAULT_ADMIN role has a second holder beside the timelock: Safe 0xa0fb1b11cca5871fb0225b64308e249b97804e99, a 6-of-9 whose owners are the same nine accounts, granted on 2025-04-02 and untouched by the April 2026 change. Its reach is bounded — it can change the treasury and harvester addresses, the expense ratio and the claim-for role (a simulated updateTreasury to an arbitrary address succeeds from it) but not principal or the implementation. Below the wrapper, the contracts fxSAVE’s value passes through carry undelayed emergency authority: the main Safe and a 3-of-4 Safe (0x28c921adac4c1072658eb01a28da06b5f651ef62, owners a subset of the nine) hold emergency roles on the PoolManager, both collateral pools and the Stability Pool, including updatePriceOracle with no validation; the deployer key 0xa1d0027ca4c0cb79f9403d06a29470abc7b0a468 alone still holds EMERGENCY_ROLE on the WBTC pool that carries 89.7% of fxUSD debt; and both collateral oracles are owned by the main Safe directly, with the transfer to the timelock scheduled on 2026-04-20 and never executed. The oracle decides what gets liquidated into the Stability Pool that fxSAVE holds. Changing what the code reads is instant; changing the code takes three days.

Custody is segregated: assets sit in a bare 45-byte minimal proxy with no admin of its own, and its gauge balance reconciled to totalAssets() exactly. A bug bounty is published at docs.aladdin.club, paying up to $500,000 for critical findings. Two audits are published, and the one serious finding on record — a double-flash-loan access-control bypass in the router peripheral, found by ChainSecurity in April 2025 with over $2M at risk — was responsibly disclosed, fixed, and never exploited.

Risks and limitations. Three contracts sit between the share and the Stability Pool, none visible through the ERC-4626 interface, and the audited perimeter stops at the first of them. The fxSAVE implementation is byte-identical to the May 2025 OpenZeppelin audited commit, and the fxUSD implementation differs from it by three lines matching a later published patch; but the PoolManager, both collateral pools, the Stability Pool and the PoolConfiguration contract were deployed between October 2025 and April 2026 and differ from the audited commit by roughly 1,500 lines. SECBIT’s March 2026 report names two of them but scopes itself to a deployment on a different chain, so no published audit covers this rework — and those four contracts custody the collateral, run the liquidation logic, hold the Stability Pool that is fxSAVE’s entire asset, and determine whether fxUSD redemption is open. That is a statement about coverage, not about a defect: private review may exist, and nothing here says the code is wrong. The oracle path compounds with an audit finding — OpenZeppelin’s H-02, price manipulation through a single low-TVL spot source, is acknowledged and unresolved, and both live oracles are non-upgradeable contracts from March 2025 carrying the pre-fix formula. The same ProxyAdmin, timelock and emergency keys govern fxUSD, so the share and the stablecoin beneath it are one administrative leg, and nine key-holders with no external governance check remain the ultimate authority. f(x)‘s public v2 deployment manifest (AladdinDAO/fx-protocol-contracts) names the ProxyAdmin and the base pool at the addresses measured here, and is silent on the TimelockController, the proposer Safe and the fxSAVE token itself — the authority structure above the ProxyAdmin is verifiable on-chain but not documented by the issuer. Chains read: Ethereum. Re-check by 2026-12-22, or on a new deployment of any of the four contracts, a published audit naming the rework, or the Stability Pool’s redemption rate widening below par. Arbitrum, Base, Optimism and BSC carry no code at the fxSAVE address; Polygon was not read; and a same-address check cannot rule out a different-address deployment elsewhere.

6 · Issuer — 5.5

f(x) Protocol is built and operated by AladdinDAO and has run on Ethereum since 2023. Protocol TVL was $134.3M on 2026-09-22, Ethereum only; two audits are published and a self-hosted bug bounty pays up to $500,000. When ChainSecurity disclosed the router vulnerability in April 2025, the team fixed it under responsible disclosure and published the report on 2025-06-17; the distinction between a vulnerability that existed and funds that were lost is worth keeping, because the first is often quoted as the second. fxUSD marked $0.99966 on 2026-09-22 and no depeg event surfaced.

Risks and limitations. This is a small DAO with no external check on its governance, operating a complex multi-contract system with no regulated wrapper. The disclosure response is evidence of engagement, not of the absence of further issues.

Who this is for

• Depositors who understand fxSAVE as a paid position in f(x)‘s liquidation backstop — a claim that earns in calm markets and converts into liquidated collateral in stressed ones — and who expect to exit through redemption into fxUSD rather than through a spot market.
• USDC lenders into the Morpho Blue markets that accept fxSAVE and PT-fxSAVE as collateral, for whom the number that matters is not the price but what an instant redemption realises. On 2026-09-22 one fxSAVE redeemed to 1.0712 fxUSD (1.108142 × 0.966643), about $1.0708 against a $1.1189 mark — a gap of roughly 4.3%. At 86% LTV that gap alone consumes about a third of the 14% cushion before oracle lag, liquidation incentive or slippage; at 91.5% it consumes more than half.

Who should avoid it

• Anyone who reads “saved fxUSD” as a stablecoin savings product. The underlying is a loss-absorbing pool, not a reserve.
• Anyone who needs a dollar exit at size. Redemption is uncapped, but it terminates in fxUSD, whose market yields at most about $4.65M of USDC at any size.
• Anyone counting fxSAVE and fxUSD as two separate exposures. They share a ProxyAdmin, a timelock and the same undelayed emergency keys.
• Anyone who reads the 72-hour timelock as covering the collateral. It covers the code; the price feed that decides what gets liquidated into the Stability Pool can be repointed without delay, and on the WBTC pool by one key.

What to watch

• The fxSP redemption rate against par (0.976407 patient, 0.966643 instant on 2026-09-22). A widening discount is the pool absorbing losses; whether the current 2.4% is fee or loss is the open question that decides how to read it.
• The gap between fxSAVE’s quoted price and its instant-realisable value, currently about 4.3%. This is the figure a Morpho lender’s cushion has to survive.
• The USDC ceiling on the fxUSD sell ladder ($4.65M peak proceeds on 2026-09-22) and the USDC/fxUSD Curve pool’s balance.
• Morpho Blue’s fxSAVE balance (18,751,881.70, 27.0% of supply) relative to that pool.
• MinDelayChange on the timelock at 0x68863fb8855b04509a835082478D6E3D0bE4E61a, any change to the proposer Safe’s 6-of-9 owner set or threshold, and any implementation change on the fxSAVE or fxUSD proxies — one queue governs both.
• Any updatePriceOracle call on either collateral pool; whether the deployer key is ever removed from EMERGENCY_ROLE on the WBTC pool; and whether the oracles’ pending transfer to the timelock is ever accepted.
• The vault’s gauge balance against totalAssets(); they matched to the wei and should continue to.

Revision history

2026-09-22 (latest) — Contract & Admin 5.0 → 4.0; overall held at 4.5. A second DEFAULT_ADMIN holder on the fxSAVE contract (a 6-of-9 Safe with the same owners, bounded to treasury, harvester, expense ratio and claim-for) read alongside the timelock; undelayed emergency authority on the contracts beneath the wrapper, including a single deployer key on the WBTC pool and unvalidated oracle repointing; both collateral oracles owned directly by the main Safe. Bytecode matched: the fxSAVE implementation is byte-identical to the audited commit; the collateral machinery is not covered by any published audit. Bounty confirmed at $500,000.

2026-09-22 (later) — Liquidity & Exit 4.0 → 3.5, overall held at 4.5. The fxUSD → USDC leg measured: proceeds peak at $4,648,576 USDC, 2% crossing at $4.25M–$4.30M, 33 Curve pools enumerated and no Uniswap, SushiSwap or Balancer venue; fxUSD’s own redemption confirmed refused, so the leg is DEX-only; 90.55% of fxUSD locked in the Stability Pool.

2026-09-22 — first staged publication. Custody chain, share supply, totalAssets, convertToAssets, the Morpho Blue balance, the proxy implementation and admin slots, the ProxyAdmin owner, the timelock delay and its role holders, and the proposer Safe’s owners and threshold read on-chain; redemption paths established by eth_call simulation with a burn-address control; venue TVL from Curve’s stable-ng factory listing. Admin addresses reconciled against the issuer’s v2 deployment manifest and the OpenZeppelin audit scope. Not done at first publication: a depth ladder on fxUSD/USDC, separation of fxSP’s sub-par redemption into fee and loss, matching the deployed implementation to the audited commit, and a Polygon read.