TL;DR

How it works

USDC deposits mint fLiteUSD and park in Fluid’s fUSDC — the whole withdrawal buffer, $14.37M (34.8% of posted NAV). A Strategy Handler (0x066e…A965) pulls the rest to per-chain strategies running Instadapp DSAs, which borrow USDC/USDT against yield-bearing stables in Fluid’s own vaults. Exit is redeem() less 5bp, reverting InsufficientIdleFunds once the buffer is short — no queue.

No oracle prices this vault and the holder’s position cannot be liquidated. totalAssets() is exchangePrice × totalSupply — a liability restated, not a measurement. The loops are liquidated downstream on Fluid’s feeds, but nothing propagates back; the accrual formula is the only live path to the posted price.

Yield — two legs, neither organic. A 6.00% fixed rate (cap 30%) and a $3M/yr reward budget (cap $100M); holders get the higher. At $41.3M the budget binds at 7.26%, so nothing credited is earned. The legs cross at $3M ÷ 6% = $50M TVL, above which accrual reverts to the base. The idle third earns 5.16% — a 2.1pp drag, and the reason to hold less buffer.

Label vs exposure. The page promises “an optimal mix of yield bearing stablecoins such as sUSDe, syrupUSDC, syrupUSDT, and sUSDai.” On-chain 95.7% of gross collateral ($145.8M of $152.4M) is sUSDai-collateralised at 83.9% blended LTV; the syrupUSDC, syrupUSDT and sUSDe vaults hold dust, and $6.55M gross is USDe and PayFi Strategy Token. An unmarketed Solana leg was added 2026-08-10.

Capital quality: rented — ~860 holders, top holder 6.3%, but supply fell 43.3% from its 2026-04-15 peak to 04-24, still 89.4% of peak.

Risks

Verified on-chain at Ethereum block 25,739,860 (2026-08-12 15:37 UTC), with Arbitrum and Plasma read at their chain heads within ~2h: price mechanics, both rate legs, buffer history from launch, pause flags, every privileged key’s composition, both proxy admins, all four strategy registrations, every funded position’s collateral / debt / LTV, sUSDai concentration, the assets-vs-liabilities reconciliation, and that fundReserves() has never emitted. Taken on trust: what backs sUSDai off-chain, and Fluid’s core code — audited, but not re-derived, and the Code4rena results are private.

IDRiskMechanismSource + our delta
R1A loss can never mark down; late exiters eat itupdateExchangePrice() reverts on any decrease and accrual is a set rate, so a loss drains the buffer, not the price — early redeemers leave at par. Assets $41.18M vs liabilities $41.34M: short ~$161K (0.39%). Exit: any InsufficientIdleFunds revert.The RFC’s headline protection is a “500,000 USDC reserve fund [that] prevents exchange rate decreases”; BA Labs calls it “seeded at launch”. On-chain fundReserves() has never emitted, getReserves() = 0, and the handler holds $0.22.
R2Reconciliation structurally broken since launch_getTotalNetAssets() reverts NavTimestampsOutOfRange past a 1h tolerance. The Arbitrum and Plasma mirrors read $3.00 / $3.01, navTimestamp 2026-03-09 — 155.7d stale — so reconcileReserves() can never run. Exit: getLastReconciliationTimestamp() still 0 next review.Ours; not raised by BA Labs. The capital is real ($6.34M + $2.44M) — the accounting, not the money, is missing. The Solana leg registers chainId = 1, so the freshness guard never applies to it.
R3Exit has already failed twice, for 15 daysRedemption is gated on getIdleBalance(), which sat below 1% of NAV on 15 of 147 daily samples: 04-21→04-24 at $0.08–$22 against ~$26.5M of claims, and 06-04→06-14 under $400K. Exit: buffer < 10% of NAV for > 24h.BA Labs got as far as “the worst-case exit time cannot be quantified precisely from public materials alone”; we quantified it from chain state. Corroboration: across the three April days gross deposits and withdrawals match within $6 and supply froze at 26,187,723 — exits were funded by that day’s depositors.
R4One un-timelocked 7-of-14 wallet can take the fundsThe 24h timelock gates the vault’s owner functions but not the proxy admins that move funds — getAdmin() on the handler and the mainnet strategy is the Avocado. addImplementation() registers arbitrary delegatecall; pullFunds() is uncapped. Exit: unexpected LogSetImplementation / LogSetAdmin.Fluid’s risk docs name a Team Multi-Sig (0xa8c3…CCC5) that “does not have access to funds” — true but irrelevant here: it is 4-of-8 and isAuth = false on this vault. The wallet that can move funds is undisclosed. Ours.
R5The whole book is one young, thinly-held credit assetsUSDai backs 95.7% of collateral and exits via a FIFO queue on 30-day epochs that roll when unfilled. 97.0% of Ethereum’s sUSDai sits in Fluid, ≥75% of it this vault — deleveraging means selling to itself. Three of ten positions sit within 0.6pp of their cap. Exit: sUSDai discount > 100bp.USD.AI’s wrapper over a PYUSD sleeve plus GPU-backed compute loans; Cayman, unlicensed. BA Labs flagged the concentration in April. Pharos records direct holder-balance freeze/seizure authority. Our delta: the $100M facility (2026-07-06, acc. 2026-08-12) is Fluid’s own — collateral, debt and buffer share one balance sheet.
R6No audit covers the deployed code, and it is 5 months oldFluid’s audits page lists seven reports covering the core Vault, DEX and Liquidity Layer. None covers liteVault, the Strategy Handler, the per-chain strategies or the bridge aggregator. Live 2026-03-08; the 24h timelock arrived only 2026-07-04.BA Labs reached the same conclusion in April, and adds a Code4rena engagement (Aug–Sep 2025) absent from the page — results private — plus a $500K Immunefi bounty. Partial mitigant: updateExchangePrice() is capped at +2% per call.

Key-compromise threshold

The vault carries one role — the Avocado multisig 0x4F6F…D49e (7 of 14 signers; one an EIP-7702-delegated EOA) — which is also its secondary owner, the timelock’s proposer and canceller, and the un-timelocked proxy admin of the Strategy Handler and mainnet strategy. The handler carries five more, the lowest a bare EOA.

Liquidity & exit

redeem() less 5bp, capped by getIdleBalance() = $14,368,663 (34.8% of NAV). No tradeable secondary market for fLiteUSD, and it is not collateral anywhere. The buffer is itself fUSDC, competing with the loops’ own borrowing inside Fluid’s Liquidity layer, which holds $24.4M free — redeeming it consumes 59% of that while the strategy is short the same pool.

Defensible size (post-haircut): $0. A deliberate override of the rubric, not its output: the rubric prices slippage and fails a venue that cannot clear in 30d, and the worst closure here was 11 days. But redemptions clear at par or revert, so the failure mode is a capacity outage — and capacity has twice been ~$0. The $14.37M is discretionary: no code enforces a floor, a 2-of-2 key can move it, and holding it costs ~73bp a year in foregone spread.

Stress scenario (“NOCA exits 50% in 7d while peers also exit”): April is the replay — $18.59M withdrawn over three days against $199,763 of deposits, buffer 39.9% → 24.1% → 7.4% → $0.08, then four days at zero where the only exit liquidity was that day’s depositors.

External reviews

One narrative assessment exists (BA Labs); two algorithmic ratings sit on the Trading Strategy page. Nothing from LlamaRisk, Chaos Labs, Bluechip, DeFiScan or Exponential. Per rubric.md, >90d is stale (⚠️) — the 180d line is an ETH-LST carve-out and does not apply here.