Borrowers

Liquidated on a wick? Here a drop has to hold for eight hours first.

Cross your line inside 4% and nothing sells for eight hours. The reserve, MBRN stakers and both lender tranches take a loss before you do.

Tested on a real day

Replayed on 10 Oct 2025 against the same oracle rounds Aave liquidated on, the eight hours inside the band close 36% less debt than Aave did across the 748 accounts whose collateral the day’s feeds can price. Aave’s own close factor repays partially too; the eight hours and the recall are Membrane’s.

Reads your open position on leading Ethereum money markets. Replays it through the measured 10 Oct 2025 prices.

One day. Same accounts.

10 Oct 2025 · 2,350 liquidated accounts

$67M protected

from forced debt closure

Debt closed across the measured Oct. 10 cohort. Lower is less forced repayment.

4% sounds small. It would have kept $1.2B of collateral over 3.6 years of Aave V3.

$67M of debt protected from forced closure on 10 Oct 2025 alone.

4%.

You aren't liquidated when you cross the LLTV. Staying within 4% of it delays your liquidation for 8 hours. You get automatic protection from wicks and time to manage your debt.

Past 4% the partial liquidation is immediate down to the maximum borrowable LTV.

01 / The waterfall

Membrane

  1. 01

    Reserve

    empty at launch

    protocol

    drained first. Empty at launch: reserve_ratio is deployed at 0.

    source

  2. 02

    MBRN under the asset

    LTV voters

    revenue stops, then their stake is sold at auction, up to the stake under that asset.

    source

  3. 03

    Junior lenders

    junior tranche

    yield burned, then capital.

    source

  4. 04

    Senior lenders

    senior tranche

    yield burned, then capital, then a haircut that later revenue refills when there is any.

    source

    5. 05

    You

    borrower

    No other borrower’s debt or collateral is touched. The cascade never writes to a position.

    source

Pooled lending

  1. 01

    Borrower

    you

    liquidated at the threshold, same block, no window.

    source

  2. 02

    Safety module

    stakers

    if one exists, and only after a vote.

    source

  3. 03

    Suppliers

    lenders

    shortfall socialised across the pool.

    source

Bond coverage

curator bonds posted ÷ the redemption volume they could fail to serve

Bonds are slashed for failing to serve a redemption: min(shortfall, bond ÷ 10) per incident. The slash is booked as protocol revenue; the unserved redeemer holds an open claim.

source

coverage · redemption basis

—

mock · registry address pending

Redemption basis

totalBonded ÷ Σ reportedAum

how much redemption demand curators can fail to serve before the bond runs out

reportedAum is set by each vault (self-attested).

source

Depth basis

totalBonded ÷ (Σ reportedAum − Σ instantLiquidity)

how big a venue freeze curators can cover from bond

instantLiquidity is a real view; reportedAum is still self-attested.

source

02 / What follows

Redemption never touches a borrower. Curator vaults deliver the served asset; no position is read or written.

Proof:

RedemptionEngine.redeem
source

Nobody else’s loss lands on you. Reserve, MBRN and both lender tranches absorb it; the cascade never writes to a position.

Proof:

Cdp._absorbBadDebt
source

A breach gets 8 hours to cure while it stays inside the asset’s band. Launch collateral is set to 4%. Past the band, the sale is immediate.

Proof:

run your position through 10 Oct 2025
source

03 / Counter-positioning

Aave

Every breach gets a cure window, and the window’s length sits behind a 14-day timelock.

why they will not copy it

Aave has no cure window to offer. Adding one makes the borrower’s time a fixed term instead of a parameter the DAO prices.

source

Morpho

Curators post a bond and are slashed when they fail to serve a redemption.

why they will not copy it

Morpho’s distribution runs through supply-side partners. Telling curators they are bonded and junior is a business-model concession.

source

Liquity

Redemption is served by curator vaults. It never opens a borrower’s position.

why they will not copy it

Redeeming against borrower collateral IS Liquity’s peg defence. Removing it removes the peg.

source

Recall

The engine pulls deployed CDT back from venues and burns it against debt inside the unwind.

why they will not copy it

To copy it Aave would have to become a yield router and own venue liability, abandoning neutral base-layer positioning.

source

Counter-positioning: each is a thing the incumbent could build and will not, because it costs them their model.

Run it against the measured day

04 / Carry

The debt earns the yield. That is what makes a recall possible.

When a position must unwind, the engine pulls the deployed CDT back and burns it against the debt. The recall itself charges nothing.

Borrow cost comes out of the carry yield. If the spread inverts, curators cover 14 days of yield to give you time to act.

The debt is deployed to venues and the yield accrues against it.

Unwinds run in the engine. No outside keeper has to fire for a recall to happen.

A recall pulls CDT back from the venue and burns it against the debt, measured by balance delta.

No protocol fee on liquidation. The liquidator’s fee is sized on what is repaid and capped by collateral value.

sourceOpen the carry page

05 / CDT

CDT is the mechanism: the debt you mint is the capital that earns, and the thing a recall burns.

Mint CDT

06 / Counterfactual · 10 Oct 2025

What this order did to 2,350 real liquidations.

Accounts

748

real, liquidated, priced

Aave closed

$34.7M

entire episode

Membrane would close

$22.2M

8h window, walked minute by minute

Difference

$12.5M

36.02% less debt closed

Same four numbers, unpriced collateral added back

Accounts

1,751

Aave closed

$47.5M

Membrane would close

$54.2M

Difference

$-6.7M (-14.13%)

1,003 of these accounts hold collateral this dataset cannot price, so they take one repay and can never cure — they carry the repay-to-cap upper bound with none of the 8-hour window it is being compared against.

Median share of the account’s debt closed

Aave — whole multi-hit episode

50.8%

Membrane — the 8h window, walked

49.5%

Aave’s side is its entire episode, every repeat liquidation included. Membrane’s is the 8-hour window walked minute by minute against the same oracle rounds, with the sale re-arming after each repay to cap.

Membrane closes less

347

46.4% of accounts — the borrower keeps more of the position

Membrane closes more

401

53.6% of accounts — Aave declined to fully close these; Membrane’s formula would have. Filter the cohort by “Membrane worse” to read them.

By collateral asset

Asset

Accounts

Aave median

Membrane median

Cured

i

WETH

435

64.4%

100.0%

73.56%

OP

353

99.8%

100.0%

n/a

ARB

235

99.9%

100.0%

n/a

LINK

191

57.1%

43.5%

n/a

AAVE

147

65.1%

100.0%

n/a

WBTC

115

50.0%

34.1%

85.22%

cbBTC

60

49.9%

30.8%

96.67%

USDC

52

65.2%

100.0%

0%

wstETH

38

50.0%

38.9%

89.47%

UNI

33

97.1%

61.5%

n/a

USDT

21

54.1%

76.3%

0%

Assets with fewer than 20 accounts are omitted. “Cured” is blank where the Oct 10 oracle series cannot price that collateral.

What the number contains

  1. 01

    Two sets, never folded. 748 accounts whose collateral the day’s feeds price minute by minute: Aave closed $34.7M, Membrane $22.2M, 36.0% less. 1,003 accounts on collateral they cannot price (AAVE, LINK, OP, ARB, UNI, CRV): the engine can only repay to cap once, and with them added in Membrane closes 14.1% more. The lens above shows both.

  2. 02

    What Membrane’s figure is: a breach inside the 4% band waits eight hours; a return under the line clears the timer and a later breach starts a fresh one; a move past line × 1.04, or still over the line at expiry, repays to the borrow cap at that minute’s price; the position then re-arms. Aave’s figure is every hit on the account across 10 and 11 Oct, measured.

  3. 03

    Five wallets are 75% of the $12.5M difference. Without the largest it is 30% less; without the five, 14.6% less. The sign holds; the size is a handful of whales.

  4. 04

    133 wallets held the whole day, $110.7M of debt. One is 58.6% of that: a $64.9M wstETH loan Aave closed 4.4% of.

  5. 05

    Under $2,000 of debt both venues close the loan whole: 932 accounts, Aave a median 99.9%, Membrane 100%; Aave’s own leftover rule and Membrane’s $2,000 minimum agree, and both exist for the same reason: below it an Ethereum liquidation costs more in gas than it recovers, so the contract indexes the floor to the gas price rather than to a choice of ours. Between $2,000 and $4,000 the minimum still closes the loan where Aave took half: 161 accounts, 150 closed whole. From $4,000 up the median account loses 49.8% of the loan to Aave and 32.4% to Membrane, across 658 accounts.

  6. 06

    290 of the 748 lose more on Membrane and 111 come out equal. 184 of the 290 are under the $2,000 minimum and total $18,050. The other 106 are past the band or sold at expiry, where repay to cap restores the line in one pass while Aave’s liquidators took a median 46% bite: $6.6M.

  7. 07

    599 accounts, $96.8M of what Aave closed, are left out of both sides: 350 on L2s whose oracle this dataset lacks, 168 healthy by Aave’s own read the block before, 80 on unpriced collateral. No breach can be located, so no window can start.

  8. 08

    Stress it, priced set: cap each repay at the largest single Aave repay on that asset that day, $20.8M; price the day from Aave’s own snapshot instead of the round log, $17.5M; put wstETH on the market stETH feed Aave does not use, $54.3M. Every one keeps Membrane under Aave’s $34.7M except the last, and that feed is the wrong one.

  9. 09

    The Chainlink aggregator in this dataset is not the one Aave read that day: none of Aave’s own prices match a round in it (median gap 0.6%, worst 5.7%). Aave’s side is priced by Aave; Membrane’s walk uses the log as a ratio path. The Aave-anchored line above is the same model on Aave’s prices.

source
Run yours through the same dayEvery account, filterable