Audit

33Labs, an independent security research group, reviewed the source of Recoup's lender pool and the contracts wired beside it, 7-26 September 2026. It made 13 findings, none rated critical. 11 are fixed, each fix verified by the reviewers, and the other 2 are accepted risks, set out in full below.

PDFFull report2.87 MB · sha256 83d9…562a

The lender pool, where lenders' money sits, is independently audited.

WHAT WAS REVIEWED

Six source files of the public repository thedelph/recoup-contracts at commit b66023d: the lender pool's accounting and withdrawals, the protocol wiring, the liquidity source, fee routing, configuration bounds and loan-to-value maths. The fixes were reviewed on main, from f6893cb through 566a9eb.

By 0x23r0 and PhantomOz of 33Labs ↗.

7 Sep · startb66023d 566a9eb26 Sep · end

WHY THE LENDER POOL

Lenders' USDC is held and accounted for in the lender pool, and an independent audit is the gate in front of any outside lender's money. So the review went where that money sits: the pool itself, and the contracts wired beside it.

The pool is open to outside lenders up to a deposit cap, which Recoup's governance timelock sets by hand, each change public for 48 hours before it takes effect. See Lend for where it stands.

WHAT IT FOUND · 13 FINDINGS, NONE RATED CRITICAL

13 found. 11 fixed.

None rated critical. 2 accepted as risks, set out in full below.

High4 · all fixed

Medium6 · all fixed

Low3 · 1 fixed, 2 accepted

FIXED · 11
ACCEPTED RISK · 2

Statuses are the report's own: its remediation review classified 11 findings as fixed and 2 as acknowledged and accepted risk. The report judged them on the code, the regression tests, the public risk disclosures and the maintainer and reviewer discussion, not on whether a GitHub issue is open. Pick a group of blocks to filter the findings.

FINAL REPORT · 29 SEP 2026SHA-256 83d9…562aCHECK THE HASH, THEN READ IT

FINDINGS · ALL 13

RATED CRITICAL: None
MAINTAINER-REPORTED: 2
PUBLIC GITHUB ISSUES: 13

TABLE VIEW · 13 rows

The 13 findings in the 33Labs report, with severity, status and the public issue for each
ID	FINDING	SEVERITY	STATUS	ISSUE
H-01	Borrower-keyed recovery provenance redirects an earlier workout's recoveryHighFixed	High	Fixed	#45
H-02	Auction replacement orphans a closed workout's recovery legHighFixed	High	Fixed	#46
H-03	Repeated request service converts more than a requester's pro-rata cash into senior claimsHighFixed	High	Fixed	#47
H-04	Manager migration strands post-close loss recoveries with no unblocked repair pathHighFixed	High	Fixed	#53
M-01	Permissionless yield sweep can confiscate an open workout's borrower yieldMediumFixed	Medium	Fixed	#48
M-02	Realised open-workout yield bypasses the dedicated sweep protectionMediumFixed	Medium	Fixed	#49
M-03	First clean workout close captures shared residual yieldMediumFixed	Medium	Fixed	#50
M-04	Uncapped stream duration plus gross entry pricing lets a timed flush overcharge new lendersMediumFixed	Medium	Fixed	#51
M-05	A lender-yield backlog above the deposit-cap ceiling can never be deliveredMediumFixed	Medium	Fixed	#52
M-06	A request serviced down to one share-wei keeps the rest of its cash floor (maintainer-reported)MediumFixed	Medium	Fixed	#64
L-01	Permissionless settle discards a borrower's sub-unit yield accrualLowFixed	Low	Fixed	#54
L-02	After a raw cash loss, the request floors can lock the whole remaining cash until a repayment or a cancelLowAccepted risk	Low	Accepted risk	#61
L-03	A paused or blacklisting USDC shuts every bond door, because the farm pays its pending USDC inside the same call (maintainer-reported)LowAccepted risk	Low	Accepted risk	#68

Issue numbers are on the public repository's tracker, where each finding was filed and answered.

What the reviewers verified for each fix

For each of the 11 fixed findings, the report names what the reviewers verified and on which commit:

H-01the fix, on 68c0c26
H-02the fix, on 68c0c26
H-03the original High-severity routes, on 68c0c26
H-04the migration-recovery regression, on 68c0c26
M-01the fix, on 68c0c26
M-02the fix and its regression coverage, on 68c0c26
M-03the fix, on 68c0c26
M-04the capped 30-day stream and newcomer full-value recovery, on 29af396
M-05successive partial backlog delivery and the preserved below-ceiling anti-capture behaviour, on 29af396
M-0672 of 72 fix-focused tests and 104 of 104 #61-preservation tests passing, on 35f0a58
L-01that repeated non-count-changing settlement preserves the same 41,254-base-unit accrual as a single settlement, on 29af396

L-02 and L-03 are accepted risks, with no fix to verify.

THE TWO ACCEPTED RISKS

These two are accepted rather than fixed. Both are disclosed in full here and in the public repository, and if either happens, lenders and borrowers carry it.

L-02 · Low#61
After a loss of the pool's own cash, queued withdrawal floors can lock all the remaining cash until a borrower repays or a floor holder cancels, with no guaranteed time. Kept by design: writing the floors down instead would make every queued lender permanently lose part of the amount she was quoted.
L-03 · Low#68
If Circle pauses USDC or blacklists the custody adapter while any farm reward is pending, bond deposits, withdrawals and harvests fail for as long as it lasts, because DexFi's farm pays out pending USDC inside the same call. Liquidation clocks keep running meanwhile. Rated Low because no one in the protocol can cause a Circle pause or blacklist.

SCOPE AND LIMITS

AUDITEDLenderPool
AUDITEDCreditWiring
AUDITEDTreasuryLiquiditySource
AUDITEDProtocolFeeSplitter
AUDITEDConfig
AUDITEDLtvMath
OUT OF SCOPECollateralVault
OUT OF SCOPECreditManager
OUT OF SCOPELiquidationAuction
OUT OF SCOPEEpochHarvester
OUT OF SCOPENAVOracle
OUT OF SCOPERiskParams
OUT OF SCOPEDirectCallAdapter
OUT OF SCOPEMintAttemptReceiver
OUT OF SCOPEReferralRegistry
OUT OF SCOPEBondDepositConsentGate
OUT OF SCOPEAdapterConsentGate

The other 11 contracts, which handle borrowing, bond custody, pricing, harvests, auctions and the joint consent some governance calls need, were outside this review. Some findings trace paths from the pool into CreditManager and LiquidationAuction, and some fixes landed there, so the reviewers read parts of those two; this page still does not call them audited.
The audit is a review of source code, not of any deployment. The audited contracts on mainnet were deployed from the reviewed source, fixes included.
A review cannot prove there are no bugs left. The report says so itself, quoted below.

THE REPORT'S DISCLAIMER

“A smart contract security review cannot prove the complete absence of vulnerabilities. This assessment was bounded by the agreed scope, review period, available documentation, and referenced commits. Additional review, deployment verification, monitoring, and a public bug bounty are recommended before production use.”

THE REPORT'S CONCLUSION

“M-06 is fixed: its independently reviewed remediation was merged into main through PR #69. L-02 is the accepted post-raw-loss withdrawal-floor lock, and L-03 documents the accepted external paused/blacklisted-USDC custody dependency. Before accepting third-party capital, 33Labs recommends retaining M-06's regression and #61-preservation coverage, publishing the L-03 incident runbook, retaining the activation gate, and verifying deployment/source parity.”

From the report's conclusion. #61 is L-02's public issue.

THE FULL REPORT

Download the report (PDF, 2.87 MB) - “Recoup Security Audit Report”, 33Labs, September 2026.

file    33labs-recoup-final-report-2026-09-29.pdf
bytes   2,870,881
sha256  83d9b929c481cf2fbf730c53b71d7f681d926fd52c58f3764e4ea7f091f0562a

CHECK YOUR COPY HEREDrop the PDF or click. Hashed locally with SHA-256.

To check your copy, run sha256sum 33labs-recoup-final-report-2026-09-29.pdf and compare the output with the hash above. Or drop it on the box, where it is hashed in your browser and never uploaded.