GHO and sGHO as Eligible Backing Assets for USDe

[Kairos_Res]

GHO and sGHO as Backing Assets for USDe

Kairos Research, for the Ethena Risk Committee. Onchain state read September 8th 2026 unless stated otherwise. Sources, contracts, methodology, and verification status are set out in the appendix.

---

Executive summary

This report examines GHO and Aave’s savings wrapper sGHO as directly held assets in USDe backing, sized at $50mm and acquired by minting GHO through Aave’s stability module with USDC.

• The credit is sound and the peg has held; the constraint is exit liquidity. Aave is overcollateralized, GHO has traded within half a cent of par for twelve months, and it did so through most of 2025 while Ethereum’s stability modules held less than $50mm on 243 of 353 days. sGHO’s vault mechanics are cleaner than most wrappers we have reviewed. Across the five stability modules Aave operates on four chains, $354mm of GHO can be minted against $68mm that can be redeemed: $27mm on Ethereum, and $41mm on Plasma via a bridge and a lending pool that is 94% utilized. Ethereum alone holds barely half the position.

• Depositing USDC creates shared exit liquidity. The current Ethereum modules received $164mm from their predecessors on March 5th 2026, grew to $241.5mm by March 22nd, and hold $27mm now, down 89% net of four separate top ups by Aave. $82mm left in the 48 hours of April 18th to 20th. The USDC module, the route this mint would use, has been effectively empty since June 12th. Plasma’s module drained in the same April window, was refilled in May, and has held within half a million dollars of $41mm for 100 days.

• Redemption flow is concentrated in a handful of execution contracts. Only thirty addresses have ever redeemed GHO through the Ethereum modules, and two routing contracts account for 74% of the $358mm. The trade they execute is mechanical: buy GHO below par, redeem it through the module at par for 10 to 15 basis points, keep the difference. Who stands behind those contracts, and whether any of them were end holders exiting, cannot be established from the chain. What can be established is that whenever inventory has existed on Ethereum it has been consumed within weeks.

Recommendation. We support holding GHO in USDe backing at up to $50mm, provided our colleagues in the Committee approve it as a position that is unwound gradually and accepts, explicitly, the risks that follow from that. The exit capacity for a gradual unwind is plausible on the evidence but not demonstrated: it rests on module throughput that only existed because Aave chose to keep refunding the modules, and on Plasma capacity that sits behind a bridge and a nearly fully lent pool. An exit inside a quarter would depend on module inventory that has not been dependable at any point in the eighteen months we can measure. The conditions in the final section are designed around that: a defined unwind schedule, a pre shortfall trigger that begins reduction before capacity is exhausted rather than after, a funded fallback, and a statement of the risks being accepted. sGHO should not sit in the Liquid Stables bucket; if the GHO is to earn the savings rate, the position should be reported as administered yield, with the exit analysis applied to the GHO underneath.

How GHO works today

Until early 2026, GHO was minted by users borrowing it against collateral and burned when they repaid. Following an architecture change in March 2026, the Aave DAO mints GHO itself through permissioned contracts and either supplies it into Aave pools as ordinary liquidity or holds it in an inventory contract the stability modules draw from. Users now borrow GHO the way they borrow any other listed asset, which pulls from existing supply rather than creating new supply.

Two consequences follow. The first is not that holders lost a claim: a GHO holder has never had recourse to any borrower’s collateral, and does not now. What changed is that GHO no longer has to be the counterpart of a collateralized borrow at all, because the DAO can mint it with no borrower behind it and place it where it chooses. Holding GHO is therefore economic exposure to Aave governance and to the treasury’s willingness to support the peg, and the DAO itself decides what stands behind the float. We do not regard that as disqualifying and Aave’s collateralization is sound, but it changes what a holder is relying on. The second is that an external party cannot mint GHO in the original sense at all. The two routes available are swapping stablecoins through a stability module at par, which is the route examined here, or borrowing against collateral.

Reported supply also overstates the traded market considerably. Measured on September 8th at Ethereum block 25,932,807:

GHO supply, September 8th 2026	Amount
Total supply	699.0mm
Held unissued in the DAO reserve on Ethereum	282.6mm
Issued	416.4mm
Idle in Aave’s Ethereum pools (Core 22.8, Prime 7.1, Horizon 25.0), supplied but not borrowed	54.9mm
Locked in the bridge, backing GHO on other chains	130.0mm
Circulating on Ethereum	231.5mm
Held in sGHO, stkGHO, and Umbrella	197.5mm
Free float on Ethereum	34.0mm

Of the GHO circulating on Ethereum, 85% sits inside Aave’s own savings products, and the free float is smaller than the proposed position. The 130.0mm bridged to other chains is not all circulating either: 59.4mm of it sits unissued in DAO inventory contracts on Monad and Plasma. Deducting that, circulating GHO across all chains is roughly 302mm. That figure is still an upper bound, because idle balances in other chains’ Aave pools are not deducted, and it is the denominator used for the cross chain percentages in this note.

The asymmetry between minting and redeeming

Aave operates five stability modules across four chains. We read all of them on September 8th 2026:

Module	Capacity to mint GHO	Capacity to redeem GHO
Ethereum, USDC	$207.4mm	effectively empty
Ethereum, USDT	$72.5mm	$27.4mm
Plasma, USDT0	$10.1mm	$40.6mm
Monad, USDC	$40.3mm	effectively empty
Arbitrum, USDC	$23.6mm	none
All modules	$353.8mm	$68.0mm

Across every module, entry capacity is five times exit capacity. On Ethereum, where the position would be held, it is ten times. The USDC module, which is the route a $50mm mint would use, has $207mm of room to accept deposits and a few dollars available to pay out. Three of the five modules, all of them USDC modules, are empty or effectively so, the two that hold anything are the USDT modules. Arbitrum’s has never held a balance in the 200 days we can see, and Monad’s has held nothing for at least the 60 days that public archive nodes reach.

What a module pays out also needs stating precisely, because it is not a stablecoin. Redeeming through a module delivers the wrapped Aave deposit token that the module holds, waEthUSDT on Ethereum and waPlaUSDT0 on Plasma. Converting that into USDT means withdrawing from the underlying Aave lending pool, which is possible only to the extent the pool holds unlent cash. On Ethereum the Core USDT pool held $215mm of cash against $2.99bn supplied, so this does not bind for a $27mm unwrap. On Plasma it does. The Plasma USDT0 pool held $38.6mm of cash against $680.5mm supplied, 94% utilized, at the block matching our Ethereum read. Ethena already holds $285.7mm of aUSDT0 in that same pool. The $40.6mm nominally redeemable on Plasma therefore competes for $38.6mm of cash with every other depositor, including Ethena’s own existing position, and the same cash cannot be counted as an exit for both.

Reaching Plasma at all means bridging GHO out of Ethereum through Chainlink CCIP, whose lane is rate limited to a bucket of 5mm GHO refilling at 1,000 GHO a second. Moving $40mm therefore takes a minimum of 9.7 hours of rate limit time, assuming a full bucket and no competing traffic, before message delivery, the redemption itself, the pool withdrawal, and the return of USDT0 to Ethereum across a second bridge. We treat Plasma as real but slow and contingent capacity, and we do not treat it as equivalent to Ethereum capacity.

The two sides are governed by different things, and this is the heart of the matter. Minting is limited by an exposure cap, which is a governance parameter Aave sets and can raise at will. Redeeming is limited by available liquidity, which is whatever stablecoin inventory happens to be sitting in the module at the moment of exit. It is not a parameter anyone controls. Minting is free, at exactly one GHO per dollar and redemption costs 10 basis points on the USDC and Plasma modules and, since September 4th 2026, 15 basis points on the Ethereum USDT module, raised from 10.

There is no per depositor reservation. The stablecoins Ethena would deposit become part of the pool every GHO holder can redeem against, and while GHO trades below par that redemption is profitable for anyone watching. Nothing stops Ethena redeeming first, but it mints in order to hold, so its exit need arises later, and the record below is what has happened to inventory in the meantime.

What the reserve history shows

GHO has had a stability module continuously since 2024, through two migrations. On March 17th 2025 Aave seized the original modules and funded replacements the same day with the same $24mm. On March 5th 2026 it did the same again, moving $164mm into the contracts that operate today. The seize function has been used twice, but both times as part of a migration rather than a withdrawal of the facility. The current contracts are in their third generation.

Over the eighteen months we can measure, Ethereum module capacity has been small far more often than it has been large. The 2025 modules started at $24mm, reached $47mm in May 2025, and fell to $5mm by August. The USDT module sat below $1mm for 161 consecutive days, from June 7th to November 14th 2025. Combined Ethereum capacity was below $50mm on 243 of the 353 days those modules operated. It grew rapidly only from December 2025, reaching $131mm at the start of January and $164mm by the March migration, and the current modules then climbed to a peak of $241.5mm on March 22nd as Aave raised caps and users sold stablecoins in. Plasma’s module was funded four days after the migration and reached $34mm by the start of April. Taken together, the modules peaked at $271.1mm on March 22nd.

They hold $68.0mm today, a fall of 75% from that peak. On Ethereum alone the fall is 89%, to $27.4mm, and that is net of active replenishment. Aave added $20.0mm to the USDC module on May 12th, rotated roughly $46mm from USDC into USDT on May 21st, added $18.3mm of USDT on July 5th, and $5.3mm more on August 12th, and reserves fell each time.

The Ethereum drawdown was not gradual. Reserves sat flat at $212mm for the week of April 18th and then $82.2mm left in 48 hours, $55.2mm of it over April 19th to 20th alone. The worst seven day window took $99.9mm, or 47.1% of the facility. Plasma was drained in the same window, from $34mm to a low of $10.3mm on April 30th. A framing of “several days of warning” is not supported by the record; the modules can lose half their depth inside a week with no observable build up.

The two chains have since diverged. On Ethereum the USDT module sat at zero for the eleven days from April 20th to 30th and was refilled, then USDC drained instead, reaching effectively zero on June 12th and staying there for the 88 days since. Plasma was refilled with $29.5mm on May 23rd and has held between $40.6mm and $41.1mm for the 100 days since June 1st. Combined capacity across the two chains has stayed above $50mm on every day since Plasma’s module opened, which is six months of history against eighteen in which Ethereum alone was usually below it.

Aave’s own stewards attribute the Ethereum drain to arbitrage against a secondary market discount that exceeded the 10 basis point redemption fee on 29 of the 30 days to August 25th. They reported on August 27th that the drain had stopped. The Ethereum modules held $36.2mm that day and hold $27.4mm now, a further fall of 24% in twelve days.

The stewards’ explanation holds in one dimension and is unproven in the other. For all but the last four days of the sample the USDT module paid $0.9990 after its fee, so any GHO bought below that was profitable to redeem; since September 4th it pays $0.9985. Over the 169 days since redemptions began, the Ethereum modules were drawn on 91% of the days the price fell more than 10 basis points below that break even, against 54% of the days it stayed at or above it. The discount governs whether the module gets tapped. In our sample the size of a draw did not scale with the depth of the discount: the heaviest days were not the cheapest days, and draw size tracked the inventory available rather than the price. That is an observation about this sample, not proof that depth has no effect on size.

At $50mm the sequence is straightforward. Ethena deposits $50mm of USDC and receives 50mm GHO. The USDC module’s redeemable balance rises from a few dollars to roughly $50mm. Every existing GHO holder can now redeem at par for 10 basis points, and while GHO trades below par that trade is profitable. GHO traded below $0.999 in 85% of four hourly observations over the 90 days to September 8th. On the record above, Ethena would then hold $50mm of GHO against Ethereum redemption capacity that had returned to roughly where it started.

Outside the modules there is not enough depth to make up the difference, and the quotes that suggest otherwise are misleading. A router will fill $40mm of GHO into USDC for about $52,000 of slippage, but it does that by sending most of the order through the Ethereum stability module itself. Strip the module out and genuine third party depth breaks down around $7.5mm. That is why $5mm of additional size costs $2.8mm. The module has fallen by $5mm since, and the ceiling falls with it. There is effectively no centralized venue: GHO is listed on Bitget and Gate, which between them traded about $3,900 a day at measurement.

On the peg itself the record has recently improved. GHO spent its first 293 days below $0.999 and reached $0.950 intraday on October 26th 2023, recovering through borrow rate increases and paid incentives rather than through a redemption mechanism. The trailing twelve months show no day below $0.995, and the lowest daily close in that period is $0.9977. That twelve months includes the second half of 2025, when Ethereum module capacity was frequently below $20mm. The peg does not appear to depend on module depth. Ethena’s ability to exit at par does, and the two should not be confused.

Who actually redeems

We reconstructed every redemption in the life of the Ethereum modules: 3,281 events, $358.3mm of GHO redeemed, at a blended realized fee of 8.5 basis points. Plasma’s module is not included in this ledger; its balance has barely moved since June.

Only thirty addresses have ever called the redemption function. Two of them account for 74%. The largest, at 38.9% and $139.4mm, first appears on April 18th, the day $82mm began leaving the modules, and has redeemed continuously since; more than 200 distinct externally owned accounts route through it, which is consistent with an aggregator or router. The second, at 35.1% and $125.7mm, ran from March 22nd to June 17th with nine accounts behind it. Third at 5.0% is CoW Protocol’s settlement contract. The top five take 86% and the top ten 95%.

A router’s 200 senders could likely be arbitrage bots, end holders leaving GHO, or both, and we have not traced beneficial ownership. We therefore do not conclude anything about motive from this concentration, and we do not claim it rules out a run. What the ledger does establish is narrower and sufficient: redemption is executed by a small number of automated paths that act within blocks whenever the arithmetic pays, and on every occasion inventory has existed on Ethereum it has been consumed within weeks.

sGHO assessed separately

sGHO is a plain ERC-4626 savings vault, and the implementation contains no slashing, cooldown, or settlement functions. There is no withdrawal queue and no exit fee, and we verified that a full redemption of the largest holder’s position, 15.6% of supply at the end of August, completes in a single transaction. On mechanics alone it is cleaner than most yield bearing stablecoin wrappers we have reviewed.

The question is what produces the return. The rate is administered rather than earned: it is stored onchain as a fixed figure, and realized return has been identical within four decimal places in every weekly window since May 19th 2026. Aave announced on August 27th that the rate would rise from 4.25% to 4.5%, and that change has since executed onchain. The vault does not generate the yield itself and the rate is not mechanically linked to protocol revenue. Yield arrives as discrete transfers from the treasury and a related multisig, 2,114,000 GHO across ten transactions since launch. Because accrual is continuous while funding is periodic, the vault has owed more than it held on 35 of the 123 days since launch, across five episodes. The longest ran for the 17 days from July 15th to July 31st, when the shortfall peaked at roughly 266,000 GHO, or 19 basis points of what the vault owed. At our September 8th read the vault was 32,152 GHO short; a 330,000 GHO top up landed 166 blocks later and left it roughly 297,000 GHO in surplus.

Control over the rate sits with a two of three multisig, which can move it anywhere between zero and 50% with no governance vote and no timelock. A separate four of seven multisig can pause the vault, which sets withdrawals to zero and blocks transfers.

Two conclusions follow. sGHO does not belong in a liquid stables allocation, because exiting it means redeeming to GHO and then facing the constraint above in full. Nor is it lending yield, because nothing is lent. It is a subsidy paid on idle balances at a rate a small group can withdraw at will. Classifying it as administered yield has three operational consequences: it is reported outside the Liquid Stables line, its yield is not projected forward in backing income because the rate can be set to zero without notice, and the exit test in this note is applied to the GHO underneath it without adjustment for the wrapper.

Aave level considerations

Aave’s internal price feed for GHO returns a fixed $1.00 with no oracle behind it and no staleness check. This is a defensible choice for an asset the protocol itself issues and is common practice, but it means Aave’s own risk systems cannot observe a GHO depeg, which is the event a reserve holder would care about most.

Aave Core also carries roughly 53,000 ETH of recorded bad debt, about $132mm at the September 8th price, the residue of the April 2026 bridge exploit and still open nearly five months later. Coverage against that deficit stood at roughly 19% at the end of August and had been declining as stakers exited. This does not impair GHO directly and Aave remains substantially overcollateralized, but it is the balance sheet standing behind the exposure.

Concentration

Whatever is decided here, the wider exposure question deserves its own answer, because the figures run larger in both directions than the headline backing allocation suggests.

As of August 31st, Ethena held approximately $505mm of Aave interest bearing tokens across four instances, roughly 40% of the DeFi Lending allocation and 12.5% of total backing, and represented 95% of one of those reserves and 91% of another. In the other direction, approximately $1.76bn of USDe and sUSDe sat within Aave, equal to roughly 43% of USDe outstanding. Aave is Ethena’s largest single source of demand.

The Plasma pool deserves a specific mention because it appears in both the existing exposure and the proposed exit route. Ethena’s $285.7mm of aUSDT0 sits in a pool that held $38.6mm of withdrawable cash on September 8th. That position is already seven times the cash available to redeem it, before any GHO is added, and any GHO redeemed through the Plasma module would draw on the same cash.

There is also a reflexive link, bounded today but set by parameter rather than structure. USDe and sUSDe cannot be used to mint GHO on Aave’s main Ethereum instance, but can on several others, where approximately $81mm of GHO borrowing was outstanding at the end of August. Against roughly 302mm of GHO circulating across all chains, that is an upper bound of about 27% of circulating GHO potentially backed by Ethena’s own assets, and a single governance proposal could widen it.

A $50mm GHO position would add a third exposure to the same protocol, governance process, and parameter setting.

The exit plan

The analysis above measures capacity at a moment. That is the right test for a position that might need to leave quickly. For a position unwound gradually the relevant question is different, and harder: not how much can be redeemed today, but how much will be redeemable, month after month, while the position is being reduced.

We understand the intended structure is a single allocation of no more than $50mm, reduced over time rather than exited at once. The only evidence we have about future capacity is past throughput. Over June to August the whole market redeemed $29mm a month through the Ethereum modules, all of it consumed by the redeemers described above, and all of it available only because Aave refunded the modules three times in that period. If that continued, the table below shows what share of it a $50mm unwind would need.

Unwind period	Ethena’s share of June to August Ethereum module throughput
3 months	57%
6 months	28%
12 months	14%
18 months	10%

This is a scenario, not a demonstration. Historical throughput is not committed future capacity, and the eighteen month record above shows Ethereum capacity spending long stretches near zero. A twelve month unwind is comfortable if Aave keeps refunding at the recent pace, but it is not comfortable if Aave stops, and Aave has not committed to anything. Plasma’s $41mm sits behind that as a second route, with the bridge and pool cash constraints already described.

A reduction rule tied to module capacity also has a limit that needs to be understood before it is relied on. If Ethena holds $50mm against $40mm of module capacity and redeems $10mm through the module, it then holds $40mm against $30mm. The ratio does not improve, because the position and the capacity are drawn from the same inventory. A trigger that begins reduction only when capacity has already fallen below the position therefore cannot repair the shortfall by itself. It has to fire earlier, while there is still a buffer, and it has to be able to use routes other than the module: the secondary market, OTC, and Plasma.

Conditions of approval, and the risks accepted

The test we would apply to a directly held reserve asset is that it can be exited at par, at full position size, within the time an exit might be needed, without depending on inventory that others will consume first. GHO does not meet that test on demonstrated capacity. It can meet it on a gradual unwind under assumptions the Committee would have to accept knowingly. We therefore recommend approval on the following conditions, and we set out the residual risks explicitly so that approval is a decision rather than a default.

1. One allocation of up to $50mm, unwound over time, with no top ups. Yield earned in sGHO counts inside the same cap and is reported as administered yield outside Liquid Stables, with the three operational consequences described above. If Ethena needs the option of exiting GHO inside a quarter, this is not the asset for that portion of backing.

2. A capacity check that starts the exit early. Ethena tracks combined redeemable capacity across the Ethereum and Plasma modules, which is public data. If it has been below the remaining position for fourteen consecutive days, Ethena begins reducing the position through whichever of the modules, the secondary market, OTC, or the Plasma route is cheapest, and reports the reduction to the Committee. On the eighteen months of history, this rule would have been in force from March 30th 2025 to November 14th 2025, when Ethereum was the only module, and has not been met on any day since Plasma opened.

3. The Plasma route rehearsed once, before it is needed. A small round trip through CCIP, the Plasma module, the pool withdrawal, and the return leg, completed within the first month of the allocation, so that the mechanics and timing are known and reported to the Committee. It shows the route works, not that $40mm can move under stress.

Risks the Committee would be accepting. As we already discussed, module capacity is discretionary, and Aave has funded, migrated, and refilled the modules at will, and has not committed to continue. Redemption inventory is shared and has historically been consumed within weeks of appearing. Plasma capacity depends on a rate limited bridge and a lending pool that is 94% lent, in which Ethena already holds a position seven times the available cash. Secondary depth is roughly $7.5mm and centralized venues are immaterial. An exit inside a quarter cannot be planned on any evidence in this note. And the paper’s throughput analysis is a projection of the recent past, not a measure of what will be there.

We would also note, for completeness, that LlamaRisk and TokenLogic’s February 2026 review of GHO, prepared for a small crvUSD pegkeeper allocation rather than for reserve use, cited stability module reserves and pooled DEX liquidity as supporting factors and conditioned its support on module depth being maintained. Both have declined materially since, though their figures and ours are not measured on the same basis, so we do not put a percentage on it. It is a reasonable point to raise with Aave directly.

---

Appendix: sources, methodology, and verification status

Figure	Source and method
Which modules exist	The GHO section of the Aave address book lists deployments on ten chains; five carry a stability module: Ethereum USDC 0x3A38…4112 and USDT 0x8822…F5E3, Plasma USDT0 0xd061…BF37, Monad USDC, and Arbitrum USDC 0x53E0…84eD. Avalanche, Base, Gnosis, Ink, Mantle, and X Layer have none. Not independently verified.
Module capacity and redeemable balances	Read from each module at the latest block on September 8th 2026, Ethereum block 25,932,807 and Plasma block 31,931,880. Caps and balances are denominated in wrapped aToken shares and converted to dollars with each wrapper’s convertToAssets. Arbitrum’s module showed no balance in 41 samples over 200 days; Monad’s showed none in the 60 days public archive nodes serve.
Reserve history	Wrapped token balance of each module at the last block before 00:00 UTC each day, converted at that day’s rate. Three generations of Ethereum contracts: 0x0d8e…3578 and 0x686F…8262 to March 17th 2025; 0xFeeb…9535 and 0x535b…6B7B from March 17th 2025 to March 5th 2026; the current contracts since. Plasma from March 8th 2026. Spot checks on five dates reproduced independently.
Migrations	getIsSeized on each retiring pair flips at block 22,065,512 (March 17th 2025) and around block 24,592,000 (March 5th 2026); the replacements held matching balances the same day.
Redemption ledger	Every BuyAsset event emitted by the two current Ethereum modules from deployment to block 25,932,819: 3,281 events. Attribution of the two largest contracts is from the sending accounts behind them, measured on September 1st; beneficial ownership is not established. Plasma redemptions are not included. Not independently verified.
GHO supply breakdown	totalSupply of GHO; balances of the DAO reserve 0x54C5…E7E9, the CCIP token pool 0x0617…660a, sGHO 0xE175…a1d, stkGHO 0x1a88…885d, and Umbrella stkGHO 0x4f82…5033; idle GHO as the token balance of the Core, Prime, and Horizon aGHO contracts; L2 unissued inventory as GHO held by the Plasma and Monad DAO inventory contracts. All at block 25,932,807 or the latest block on each chain.
Pool cash	Underlying token balance of the aToken contract: USDT0 in the Plasma pool and USDT in the Ethereum Core pool. Ethena’s aUSDT0 balance from its backing wallet 0xb873…313C.
GHO price	DefiLlama coins API for ethereum:0x40D1…6C2f, daily from July 15th 2023 and four hourly from February 19th 2026. Intraday low from the hourly series for September to December 2023. The 90 day figure is 85% of four hourly observations; on daily closes it is 82%. Principal lows reproduced independently.
Break even and discount analysis	Break even of $0.9990 from the USDT module’s 10 basis point fee, in force until 23:07 UTC on September 4th 2026, when the fee rose to 15 basis points; the module’s quote now returns $0.9985. Incidence analysis buckets each day by its four hourly low against break even and tests whether any redemption occurred. Not independently verified.
Bridge limits	getCurrentOutboundRateLimiterState on the GHO CCIP token pool; the Plasma lane is selector 9335212494177455608; all lanes carry a 5,000,000 GHO bucket refilling at 1,000 GHO a second. The 9.7 hour figure is rate limit time only.
Secondary depth and venues	Routing aggregator quotes for GHO into USDC at $40mm, $45mm, and above, taken September 1st 2026; dollar slippage derived from the quoted rates. Centralized venue volume from the two non DEX tickers on CoinGecko, Bitget and Gate, on August 31st. Not re-measured on September 8th.
sGHO rate and funding	targetRate, totalAssets, and the vault’s GHO balance sampled at the first block of each day from May 5th to September 4th 2026, and at blocks 25,932,807 and 25,932,973; funding identified as GHO transfers into the vault with no matching Deposit event in the same transaction. Deficit days and episodes reproduced independently; funding reconciliation not.
Aave bad debt	getReserveDeficit on the Aave Core pool for every listed reserve on September 8th 2026; the deficit is recorded against WETH at 52,964 ETH, priced at $2,484. Umbrella coverage is carried from August 31st.
Steward statements and prior assessments	Aave governance forum, GHO Stewards update of August 27th 2026 (topic 25534) for the 29 of 30 days figure, the drain statement, and the rate increase; LlamaRisk and TokenLogic, Pegkeeper Onboarding Review of GHO, February 27th 2026.
Ethena’s Aave exposure	Ethena’s published backing dashboard and Aave reserve data as of August 31st 2026, as set out in our supporting research file. Not re-measured on September 8th.