Closing YieldVault can burn residual AutoBalancer yield tokens (mainnet WETH↔tauUSDFv), returning less collateral on close

[liobrasil]

Summary

When closing a YieldVault backed by an FCM (FlowCreditMarket) strategy, residual yield tokens (e.g., tauUSDFv) remain in the AutoBalancer and are burned during cleanup. This occurs because FCM's withdrawal logic only pulls enough from the AutoBalancer to repay the MOET debt — any yield profit beyond that stays in the AutoBalancer and is destroyed when the YieldVault is burned.

Impact

• Value loss: User loses accrued yield profit (the residual tokens burned)
• Affected strategies: Primarily FCM-backed strategies (mUSDFStrategy) — non-FCM strategies have minimal risk (see below)

Mainnet Evidence (WETH → tauUSDFv)

Transaction	Link
Open	https://www.flowscan.io/tx/d726d3ac61a2b72de5644f7ed0c07b3d3715e8770fa02f1493c73a81526c5007
Close	https://www.flowscan.io/tx/5fff138475e00daa7ed1580669ddb5d82f4566a136f5ca1dab323699b0d45e8a

From close tx events:

AutoBalancer.Withdrawn amount=0.01911391, balanceAfter=0.00000759
FungibleToken.Burned amount=0.00000759  ← residual tauUSDFv burned
AutoBalancer.ResourceDestroyed

Root Cause

1. closeYieldVault calls yieldVault.withdraw(amount: getYieldVaultBalance()) (FlowYieldVaults.cdc:415)
2. For FCM strategies, the Strategy's source is position.createSourceWithOptions(type: collateralType, pullFromTopUpSource: true)
3. FCM's Position Source determines withdrawal based on debt repayment requirements, not total AutoBalancer balance
4. Since yield accrued, AutoBalancer holds more value than needed to repay — FCM only pulls what's necessary
5. Burner.burn(<-yieldVault) triggers Strategy.burnCallback → _cleanupAutoBalancer → residual tokens destroyed

Additional Issue: Inconsistent Quote Logic in SwapSource

There's also an inconsistency in SwapConnectors.SwapSource:

• minimumAvailable() uses quoteOut to estimate collateral from all yield tokens
• withdrawAvailable(maxAmount) uses quoteIn when maxAmount == minimumAvail due to < condition

// SwapConnectors.cdc:600-609
var quote = minimumAvail < maxAmount    // ← uses < not <=
    ? self.swapper.quoteOut(...)        // withdraws ALL (when minimumAvail < maxAmount)
    : self.swapper.quoteIn(...)         // calculates needed input (when minimumAvail >= maxAmount)

When closing PM strategies where maxAmount == minimumAvail:

• Condition minimumAvail < maxAmount is FALSE
• Uses quoteIn which may return slightly less input due to swap math rounding
• Tiny dust left in AutoBalancer → burned

Why Non-FCM Strategies Are Less Affected

For PM strategies (syWFLOWvStrategy, tauUSDFvStrategy, FUSDEVStrategy):

• No FCM debt constraint — on close, maxAmount == minimumAvail
• SwapSource uses quoteIn(forDesired: maxAmount) which should withdraw nearly all input
• Residual risk: Tiny dust from swap rounding (not significant yield loss)
• FCM strategies: Leave entire yield profit because maxAmount < minimumAvail (FCM limits withdrawal to debt repayment)

Strategy Type	Residual Risk	Amount
FCM strategies (mUSDFStrategy)	HIGH	Entire yield profit (significant)
PM strategies	LOW	Swap rounding dust (tiny)

Proposed Fix

1. Fix SwapSource quote consistency (SwapConnectors.cdc)

Change < to <= so that when minimumAvail == maxAmount, it uses quoteOut (consistent with how minimumAvailable() calculated the value):

var quote = minimumAvail <= maxAmount   // ← change < to <=
    ? self.swapper.quoteOut(forProvided: self.source.minimumAvailable(), reverse: false)
    : self.swapper.quoteIn(forDesired: maxAmount, reverse: false)

This ensures PM strategies withdraw ALL yield tokens on close, eliminating dust.

2. Sweep AutoBalancer before burn (FlowYieldVaults.cdc)

Add closeYieldVaultAndSweepAutoBalancer method that:

1. Withdraws collateral via normal flow
2. Sweeps any remaining AutoBalancer balance to a caller-provided receiver
3. Optionally swaps residual yield tokens back to collateral
4. Then burns the YieldVault

This is a safety net for FCM strategies where minimumAvail > maxAmount due to debt constraints — the SwapSource fix alone doesn't help there.

3. Update close transaction (close_yield_vault.cdc)

Use the new closeYieldVaultAndSweepAutoBalancer method and handle the swept yield tokens (swap to collateral if route exists).

[Kay-Zee]

Is this not the exact scenario for scenario 2 in our test cases? where the yield accrues and then the position is closed?

[Kay-Zee]

So i think the problem is there, but i'm not convinced the root cause is correct. For example, i think this statement is not true:

FCM's Position Source determines withdrawal based on debt repayment requirements, not total AutoBalancer balance

since for example, the scenario 2 tests have a consistent debt amount, but the final withdraw values are different, which would indicate this statement is not true.

I think the main culprit is actually the swaping inconsistencies, leaving dust, which is discussed later in the issue.

[liobrasil]

So i think the problem is there, but i'm not convinced the root cause is correct. For example, i think this statement is not true:

FCM's Position Source determines withdrawal based on debt repayment requirements, not total AutoBalancer balance

since for example, the scenario 2 tests have a consistent debt amount, but the final withdraw values are different, which would indicate this statement is not true.

I think the main culprit is actually the swaping inconsistencies, leaving dust, which is discussed later in the issue.

FCM’s PositionSource determines the withdrawable amount based on position health constraints (collateral/debt + risk params) and, when pullFromTopUpSource=true, the maximum top‑up available from the configured topUpSource (topUpSource.minimumAvailable()). During withdrawAndPull, FCM then pulls only the amount needed to restore/maintain health (i.e. requiredDeposit / idealDeposit), which means it withdraws only the corresponding yield-token input from the AutoBalancer — not the AutoBalancer’s full balance.

[liobrasil]

Is this not the exact scenario for scenario 2 in our test cases? where the yield accrues and then the position is closed?

Scenario 2 has a similar shape (yield price changes → rebalance → close), but it uses a mockSwapper with no swap fees or slippage applied.
Also, the original Scenario 2 test was mostly validating “value moves as expected” and didn’t verify the important post‑close invariants (e.g. “AutoBalancer inner vault is empty at cleanup” / “FCM pid is fully unwound with ~0 residual debt/collateral”).

I updated rebalance_scenario2_test.cdc but didn't succeed in reproducing the mainnet issue.
#154