Security & Testing — Forest Road Vault

The application identifies its active network and deployment receipt at build time. Production promotion is authorized only after the gates below are cleared. Token characterization is a matter for counsel; nothing here is a securities-law representation.

Forest Road Vault custodies capital against real legal claims, so correctness and test rigor are treated as the deliverable — not an afterthought. This page summarizes the posture; the full invariant specification and role model are published alongside it.

Testing

Foundry covers the contract set with targeted branch/revert tests, realistic end-to-end flows, stateful invariants, and fuzz campaigns.
ADR-0031 changed the exact source after the July 27–28 internal audit snapshot. Those historical 855-test / 442-function / 2,427-line figures are superseded and are not presented as current assurance. Four successive external fee-accounting reviews then found and re-checked the impaired-flow, dual-NAV, and exit-carry defects. The latest found no High issue and no safety-invariant break. Forest Road accepted the remaining protocol-revenue-only global-HWM tradeoff; the follow-up zero-vesting and hardening delta still awaits exact-source independent review. Its local, fork, coverage, static-analysis, and symbolic evidence is recorded in CURRENT_VERIFICATION.md.
A differential model cross-checks the attestation oracle, and reward accounting is reconciled against independent ghost accounting.
Halmos checks five backing-transition properties (68 paths) and the cascade arithmetic property (9 paths). The published formal-methods note states the model and token trust boundaries; this is not a whole-protocol formal-verification claim.
Mainnet-fork tests exercise the USDC-only launch surface against canonical Ethereum USDC at a pinned block, including the complete lifecycle, governance, compliance, oracle, points, queue, marked-to-market, deployment/handover, and loss-cascade paths.

The safety spec

Nine system invariants are encoded as stateful properties and exercised across fuzzed reachable states — the backing invariant (supply never exceeds backing), value conservation in the waterfall, strict loss-cascade ordering, the synchronized mint gate, FIFO redemption with no double-claim, concentration limits, exchange-rate monotonicity, access control, and reserve reconciliation. See Protocol guarantees for the full list and how each is enforced on-chain.

Review

A historical internal, multi-round adversarial audit was run against its then-current contract set — independent reviewers rotating attack lenses each round (arithmetic, economic/MEV, upgrade & storage, external-token integration, access control), plus static analysis (Slither). No Critical finding was confirmed. Every High finding and its disposition is published: deployment authorization was remediated with regression tests; the remaining curator-capital settlement risk is explicitly accepted/deferred rather than described as fixed.
All contracts use audited, standard implementations (OpenZeppelin) for tokens, ERC-4626, ERC-721, access control, proxies, reentrancy guards, and pausing — nothing security-sensitive is hand-rolled. Checks-Effects-Interactions and reentrancy guards are used throughout the value-moving paths.
Two subsequent five-pass source-level audits were then run across the production contracts and their deployment wiring, surfacing eighteen further findings at module boundaries the earlier tests had treated as trusted — backing/valuation and facility lifecycle in Round 1, and compliance, the vault, the redemption queue and the treasury in Round 2. We publish all of them, with per-finding remediation status, in the Audit Register — where every round keeps its own findings and its own remediation history. ADR-0030 subsequently removed the legacy multi-stable, DSRA, reserve-instrument, recovery-top-up deployment, and compatibility surfaces. The fresh clean-v1 stack formerly deployed at Sepolia block 11340997 is now explicitly archived because ADR-0031 and later liveness remediation changed the source. The current 7eef49b suite was deployed across blocks 11386373–11386520, explorer-verified, and exercised through all 661 callable ABI entries on a finalized fork. That deployment evidence does not close the outstanding independent source-delta review. No historical completion label or hash applies to the current source. This is exactly why an internal review is not a substitute for an external one.

An internal review is not a substitute for an independent external audit.

Production assurance gates (human-owned)

The internal engineering review does not itself authorize production. A first mainnet promotion requires an external security audit, a securities-law opinion, an executed legal wrapper, an economic review, and acceptance of the attestation-trust model.

Current status of those gates:

External security audit — CURRENT DELTA REVIEW OUTSTANDING. Four independent ADR-0031 review rounds are complete; the latest confirmed the holder-protective exit/backstop fixes and reported no High issue. It preceded the current zero-vesting, legacy-seed, upgrade-order, liveness, deployment-receipt, UI, and CI follow-up. That exact-source delta has not been independently closed, so the production gate is not satisfied. A subsequent internal adversarial audit of the whole protocol (Round 16, 2 August 2026) reviewed the current source directly and reported several live findings. Forest Road formally accepted two corrected residual risks on 3 August 2026: D7-01 at Medium and D4-01 at Low. Neither mechanism is resolved or refuted, but neither is a release blocker after acceptance. D7-01's loss-avoidance channel is closed by the existing atomic-private default procedure; its surviving risk is a throughput-cap bypass. D4-01's anonymous path is a bounded, non-compounding one-epoch delay. The round's four Low assurance-chain findings (D13-01 through D13-04) were subsequently remediated with tracked regressions, but the open protocol findings remain — so the delta this gate refers to has grown rather than closed. That round also corrected four of its own earlier conclusions, including one previously published here: the upgrade role is held only by the timelock in the current configuration, but the default administrator role administers it, so that is a configuration state and not a control. It remains an internal review and does not move this gate.
Securities-law opinion — OWNER-REPORTED COMPLETE. The repository does not independently attest the underlying letter. Nothing on this site is a securities-law representation.
Executed legal wrapper — OWNER-REPORTED COMPLETE.
Economic review — OWNER-REPORTED COMPLETE.
Attestation-trust model acceptance — OWNER-REPORTED COMPLETE.
Mainnet operator ceremony — BLOCKED. Tracked default/heavy tests and Solidity formatting pass, and the pinned-fork recovery rehearsal now exercises the real Treasury Safe proxy. The Safe shared-owner common mode is formally accepted as SAFE-CD-01 and remains disclosed rather than described as independent control. The validated KMS deployer is now funded with ETH and canonical USDC. Forest Road approved its four gas-policy values, and the strict read-only preflight passed at mainnet block 25,675,456 without a signing request; it must be repeated after source freeze and immediately before the ceremony. The roleless atomic MTM executor and private keeper worker now exist. A follow-up internal audit fixed four additional Medium findings and the Low KPA-L1 repository limitation; the hardened unit suite passes 24/24, the executor suite passes 12/12, and the complete compiled-worker lifecycle passes on a disposable fork pinned to mainnet block 25,500,000, including near-cap repeated-ID multi-bundle scheduling and both code-bearing and no-code bad-queue rejection, plus fail-closed expired-entry eviction. The local relay and guardian adapters prove worker behavior, not a production provider or actual Safe. The executor remains undeployed and externally unaudited; the selected bounded-batch feed lacks a production load receipt; and two independently controlled funded hosts, live private-relay/independent-heartbeat evidence and the actual guardian-Safe drill are absent. Human recovery timing and capital-policy records also remain open. A local test cannot mark those controls complete.

Once launched, those controls and the documented monitoring, incident-response, and governance processes remain ongoing obligations.