Audit register
Dated reviews, newest first. Each record states what was reviewed, how it was checked, its material findings and their current disposition. Large claim corpora remain available in the named full reports.
Reviews conducted by a party other than Forest Road.
An outside engagement carries weight an internal round cannot. Its stated scope and methodological limits remain part of the evidence and are published alongside the findings.
Corrovera review of the Solana curator vault
Four Medium findings covered frontend release identity, reproducible builds, test assurance and emergency halt latency. Three closed in the first package; the test-assurance work and a notice/draw regression required a second package. The final source has no open High or Medium program-source or test-assurance finding. Solana mainnet counsel, Squads and specialist-review gates remain separate.
Corrovera review of the accrual and open-finding diffs
The rounds confirmed and corrected the assessment one-second-expiry defect, the cold-gas impairment probe, armed-exit admission gaps, BSC reconciliation and pre-default legacy PIK loss. Two Medium findings were accepted only for the unused legacy upgrade route; Ethereum V2 instead deployed fresh with accrual active at genesis.
Corrovera dual-chain source ensemble
The initial corpus contained no consensus High: both proposed High claims split between reviewers and were later refuted once dependencies were visible. Thirty-five Medium claims had initial two-reviewer agreement and 31 were unresolved at publication; later triage consolidated duplicates, rejected context-free claims and corrected the confirmed defects. The large claim corpus remains in the full report rather than being restated as 317 confirmed findings.
Cantina Managed review of the supply gateway
Five findings, ALL INFORMATIONAL. No Critical, High, Medium or Low. One fixed, four acknowledged. The Cantina Managed team statement reads: 'No significant issues were identified during the assessment, and the protocol is expected to operate as intended.' READ THE SCOPE FIRST: two files, not the protocol. The fixed finding is the missing deadline on the two-argument redeem, which ADR-0034 W required and which is a free option for a searcher who holds the transaction until the ratio decays to the caller's floor; the remedy was documentary, recording which form is canonical and pointing integrators at the three-argument form. The four open ones are an overflow panic on a max-sized redemption input reachable only when under-backed, immutable module wiring that would need an upgrade to correct and is already covered by post-deploy manifest validation, a previewRedeem that omits the ReserveManager pause and so can quote a price no redemption would settle, and direct redemption not consuming the ADR-0033 exit interlock. That last one is the substantive one: the state that leaves par reachable is an ACTIVE RESERVE-LOSS ARM BEFORE ANY PHYSICAL SHORTFALL, accepted as a known limitation of mainnet v1. The report's trust-assumption section is the more useful read and states plainly that attesters are the primary trust boundary for off-chain facts: that the protocol cannot independently verify an attested event once the oracle accepts the signatures, and that SENIOR CAPITAL IS IMPAIRABLE with prevention of every senior loss NOT a protocol invariant. This review does not supersede the mainnet-v1 full audit of 16 August, whose open DV findings remain the substantive picture. The full report is published at github.com/Forest-Road-Company/forest-road-vault under audit-reports/.
Full audit of the live mainnet deployment
Supersedes the same-day deployment audit, which declared two gaps, no fork reproduction, so no finding could reach confirmed, and no independent adversarial cross-examination, and closes both. Three Medium and five Low code findings, all reproduced or verified against deployed bytecode, plus the deployment-configuration and disclosure findings carried forward. One candidate was refuted outright and one de-escalated Medium to Low; register entry F-01 was corrected DOWNWARD after its stated exploit path was found no longer to reproduce. THE FINDINGS ARE NOT FUND-THEFT-BY-ANYONE: the three Mediums are two privileged-operator liveness traps and one launch-sequencing gap, and no user funds are at risk at the current seed state. Two matter most to anyone integrating. DV-03 records that both junior cascade layers are EMPTY ON CHAIN TODAY, so a declared default drives full principal onto seniors with nothing in front of it. DV-01 produces an UPWARD discontinuity in the exit price, which any consumer recording a high-water exchange rate must understand before integrating. The full report is published at github.com/Forest-Road-Company/forest-road-vault under audit-reports/.
External review by Corrovera Security
Two new Medium findings and one challenge to an existing risk acceptance, all outside the 36 register entries supplied as exclusions. One lineage produced three candidates and the other four produced none; the strongest survived refutation by three independent reviewers, all of whom rated it High, and was rated down to Medium by the reviewing engineer on reachability grounds after Forest Road raised the 21-day request-anchored redemption cooldown. F-01 gates third-party curator onboarding rather than describing present exposure. F-02 requires no misconduct at all, which raises its likelihood rather than lowering it. NO FINDING REACHED CONFIRMED, fork reproduction was not part of the engagement, no property was formally proven, and no fuzzing beyond the project's own suite was run. The report states plainly that a clean section is evidence these methods at this depth found nothing there, and is not evidence of security. Open C-01 was confirmed still open. All 38 in-scope Slither Mediums were individually adjudicated and dismissed with reasons; one outside candidate was self-refuted by the reporting model during its own verification.
18 internal rounds, newest first.
Internal engineering reviews do not substitute for external audit. They are published on the same terms, every finding, every severity, every disposition.
Review of the exact deployed Ethereum V2 contracts
No new Critical, High or Medium contract defect was confirmed. The clean fork population passed 537 tests across 69 suites with zero failures or skips. One earlier rounding availability edge remains owner-accepted with a funded bounded keeper; one Low fee-withholding policy and one informational upgrade constraint remain disclosed.
Solana and curator correctness review
The review found two Medium interface/configuration failures and seven Low program or interface issues. The revoked-position access defect was fixed immediately; the Reown project was replaced and configured for the intended origins. The 20 September external review and remediation packages supersede the remaining source status.
First audit of the marked-to-market executor
Found because we were writing a document about where our own review method is blind: nothing in our process detects a production contract added between rounds. Six reviewers raised 44 candidate findings; two survived adversarial verification and both are Low. MTM-02 was then fixed: the keeper derives the digest, validates signatures and the available protective action locally, uses fixed reviewed gas and sends no pre-inclusion digest commitment, unsigned valuation or bearer calldata to either read RPC; the pinned-fork lifecycle enforces rpc-commitment-leak=0, rpc-valuation-leak=0 and rpc-bearer-leak=0 and rejects a deterministic no-action bundle before any relay call. Forest Road formally accepted the live, PROVEN MTM-01 risk on 4 August 2026; it is not fixed or refuted. The design's central claim: that an ordinary caller cannot choose a weaker protective action, otherwise holds, including against a timing-based downgrade attack that turns out not to exist. All 172 custom errors in the production source were enumerated for a collision with the fallback trigger: none collides. Two of the three hypotheses we seeded into the audit were refuted by it, one corrected by Forest Road against our own analysis.
Internal adversarial audit of the whole protocol
The core value-custody mechanics held under direct attack: cascade ordering, conservation and subordination headroom across 163,840 stateful calls with the per-event backstop cap binding for the first time; queue solvency, FIFO and no-double-claim across another 163,840; the whole attestation signature layer; reentrancy closed by building and installing a hostile module rather than by argument; and flash-loan atomicity tested by genuinely borrowing at a pinned block. The failures cluster in exactly two places, economic controls derived from spot-read balances, and the role-admin topology, where one missing override produces three symptoms that had been filed as three findings. The round also corrected four of its own earlier conclusions. On 3 August 2026 Forest Road accepted the corrected residual risks for D7-01 at Medium and D4-01 at Low; both mechanisms remain proven and live, but neither is resolved or a release blocker. As measured on 2 August 2026 against the then-36-file contracts/src, production source reported 100% line and branch coverage and every defect found lay in covered code. Two caveats now attach and must be read with that figure: the source has since grown to 49 files, which this measurement does not cover; and finding P-28 subsequently established that forge scores a multi-clause guard as a single branch, so a reported 100% branch figure is compatible with individual clauses never executing. Treat the number as a floor on line coverage for the 2 August tree, not as a current claim of exhaustive branch coverage.
Protocol fee stack, full-delta review
The fee stack composes correctly: the impairment ordering that guards the never-suppressible loss path holds by construction: the fee arithmetic cannot degenerate in any reachable state, and every cross-module lock is atomic. The cross-slice defect is elsewhere, the launch decision to recognize yield instantly silently voided the band clause of a vault-entry guard in code this work never touched, reducing it to the point test its own rationale rejects. Reproduced executably; the recommended fix was implemented and measured at 116 bytes against 88 bytes of remaining contract size, so it cannot currently be applied.
Protocol fee stack, instant-recognition re-check
The first round in this sequence whose substantive findings are about the governance record rather than the arithmetic. Both behavioural changes are implemented correctly and there is no High finding. Instant recognition retracts one of the two original reasons for vesting honestly and correctly, but overrides the other, a red-team finding about pro-rata capture at the payment instant, without dispositioning it, and the pre-mainnet economic-review gate is carried against a parameter value it was not signed against. One accounting finding remains open on which valuation base the fee-share mint is denominated in; its arithmetic is confirmed and its reachability is argued rather than executed.
Protocol fee stack, composition re-check
The first round in this sequence that does not continue it: no High finding, no safety-list invariant broken, and the surviving residual costs the protocol its own revenue rather than costing holders theirs. The exit fix delivers a real guarantee: the per-share high-water mark is now monotonically non-decreasing across every exit, which makes the earlier failure direction structurally impossible. The round's substantive result is an impossibility: basis-additive entry, pro-rata exit and round-trip neutrality cannot all hold with a single scalar hurdle, which is why three prior remediations each relocated the same defect. That trade-off is a financial-mechanic decision and is referred upward rather than patched.
Protocol fee stack, dual-NAV re-check
All four prior code findings were closed, but the exit leg still adjusted a performance-NAV-denominated hurdle using assets priced on the higher redemption NAV. That let an exiting holder transfer deferred fee exposure to stayers. The working tree now uses the greater of asset carry and pro-rata carry, fixes the adjacent upgrade and backstop-ordering issues, adds divergent-NAV assurance, and discloses the global deferral risk; independent review of that delta remains pending.
Protocol fee stack, remediation re-check
Three of the four Round 10 fixes were properly closed. The fourth reintroduced the same defect class on a different axis: the bracket added to make junior-capacity writes fee-neutral inferred its hurdle adjustment from an observed NAV movement rather than the capital that moved, making the high-water mark non-monotone and producing both an over-charge and a mirror under-charge. Round 12 independently confirmed that all four code findings from this report were closed, while identifying a separate dual-NAV exit defect.
Protocol fee stack, ADR-0031
Twenty-nine candidates, fifteen survivors, consolidated to four code defects and six findings about the tests. The headline defect is that the fee-free hurdle ratchet was keyed on a per-share rate while the hurdle it protects is proportional to share supply, so it was not asset-preserving during a live impairment and the eventual cure was charged as profit. The more uncomfortable half of the round was the test suite: the entire exit half of the change had no value-asserting test, and the rate-integrity invariant had been weakened rather than the code fixed.
Full-system audit, round nine
Eighty-four candidates, consolidated to thirty, of which two were dropped outright by the critic: one would have recommended undoing an earlier fix. No Critical or High finding, and nothing that reaches a loss of funds, an unauthorized mint, a backing break, a cascade inversion or a compliance bypass. About half of what survived is documentation drift. The round's clearest new defect was a residual on the previous round's own remediation, and the honest conclusion is that source review has now plateaued.
Remediation re-check
A dedicated re-audit of the previous round's fixes, on the principle that a remediation pass is itself a change that can introduce defects. It found exactly that: one of the twelve fixes had introduced a Medium liveness regression that could permanently freeze the only sUSDfr exit. That regression, and one further residual, were reproduced by executable proof before being fixed rather than argued from source. All five findings are now remediated, along with the three structural gaps the critic refused to sign off on.
Full-system multi-pass audit
A full re-audit of the whole system after the Round 6 remediations, six rounds in. Eighteen independent reviewers raised 71 candidates; these merged to 30, and every one then faced two refutation attempts. All 30 mechanisms held, but 22 had their consequence corrected downward, four filed Mediums ended as Informational. No Critical or High finding. Nothing breaks the backing invariant, inverts or skips the loss cascade, bypasses the mint gate or compliance, over-distributes the queue, or reaches an unauthorized mint. Post-audit, all two Medium and ten Low findings were remediated and verified; the remaining 18 Informational findings retain their original dispositions.
Release-branch differential review
A differential review of everything the release branch changed. Eleven candidate issues were raised and each was adversarially re-verified against source: one was refuted outright, one restated an earlier finding, and most had their consequences corrected downward. The live Medium, the attestation-boundary Low, and four frontend/documentation findings were remediated on 28 July; one grouped efficiency finding was partially remediated. No Critical or High finding, unauthorized-mint path, or permissionless theft path was confirmed.
Clean mainnet-v1 defensive audit
The controlling review for the current release. Eight findings, two High, six Medium. Four are now remediated with regression tests; four were explicitly accepted or deferred by the protocol owner for launch, including one High. Accepted is not fixed, and the prerequisite for each acceptance is recorded so it can be revisited.
Pre-mainnet campaign
The largest campaign run against the protocol, and the one that found the most serious defects, including a Critical path that burned whole queued positions for zero assets. Findings below are the consolidated register with each item's current status. The register is the authoritative published record: it carries every finding, its severity and its current disposition. The underlying working documents are superseded by it, they describe contract addresses and evidence from deployments that have since been archived and replaced, and are retained internally rather than published, because publishing several undated reports that disagree with the current register would mislead rather than inform. They are available to a reviewer on request.
Five-pass source review, Round 2
Nine further findings in the modules the first round treated as trusted. Most were resolved by later architectural decisions rather than point patches, and one Informational item was later escalated to Critical when a design change destroyed the premise that had made it immaterial.
Five-pass source review, Round 1
The first structured review. Nine findings, none of them elementary coding mistakes, they sat at the module boundaries the tests treated as trusted, principally backing valuation and the facility lifecycle.