The live V1 protocol contracts (13 types, 19 instances) match the compiled public source byte-for-byte (immutables zeroed, metadata included). Core at commit a26e0063; only Marketplace is a later incremental deploy at d5dfc730. The CAW token is a pre-existing external contract and out of scope. Even though the explorer shows them unverified, our own compile alone establishes that the running code equals the published source.

See the record →

Every mutable contract (12 Ownable instances) is directly owned by a single deployer EOA — the expected state before mainnet. Seven contracts are ownerless by design (Marketplace fixes its payment token at construction, FontData is pure data, etc.). The key’s power is limited (wiring is frozen in B-1, logic-swapping is ruled out in A-3), and the source itself plans a pre-mainnet multisig transfer / renounce. Neither flattering nor disparaging — we count this honestly as a door still open.

See the record →

Whether the logic can be swapped after deploy via a proxy. All contracts are non-proxy (EIP-1967 impl/admin slots empty, no EIP-1167 clones, the bodies run to tens of thousands of bytes). The source has no upgrade mechanism and no selfdestruct; the only delegatecall is a whitelisted dispatch where CawProfile / CawProfileL2 call their own immutable code.

See the record →

Whether A-2’s current owner value was actually reached by real transactions. Every Ownable contract has exactly one OwnershipTransferred (the 0x0 → deployer at creation), with zero transfers since. Each newOwner matches the current owner(), confirming A-2’s single owner has been so since creation. The multisig transfer / renounce has not yet happened (the transfer event stopping at one = A-2’s plan not yet executed).

See the record →

Read out every burned-in immutable: the root-of-trust LayerZero endpoint is 8/8 canonical (canonical EndpointV2), and the internal wiring (cawProfile, externalSelf, layer1EndpointId, relay.cawActions, the L1 mirror’s bypassLZ, etc.) all matches, with no drift. The live V1 CawActions is a thin implementation without capOracle / erc1271Sibling / zk, so that class of drift cannot arise by construction.

See the record →

The NFT representing the profile / username is CawProfile. The review’s “username-NFT contract review” is already covered by the verified targets in code-match (A-1), ownership (A-2), non-proxy (A-3), provenance (A-4) and wiring (A-5). Stated explicitly here to match the reviewer’s vocabulary.

See the record →

Checked on-chain whether the vector of forging a cross-chain peer to hijack routing is closed. setPeer is once-per-eid (OnlyOnce), so a wired peer cannot be re-pointed even by the owner — the dev comment itself names the threat (rogue / compromised owner), a stronger guarantee than standard OApp. Demonstrated by a differential test fired from the owner (5 wired peers revert / only the unset one succeeds). Owner powers such as setDelegate are noted alongside as an honest caveat continuous with A-2.

See the record →

Read out all 8 directed edges and confirmed every configured peer points to the intended counterpart contract (8/8 PASS). Both the profile mesh and the fraud-proof mesh close correctly as a round-trip-consistent mesh.

See the record →

The review’s “delegate proof.” Read the delegate of all 8 OApps from LayerZero EndpointV2: 8/8 have delegate = owner = deployer EOA (single, no drift). setDelegate is LayerZero’s standard onlyOwner, and the delegate handles security settings that need updating (DVN, etc.), so not fixing it is the correct design (forged messages can’t be made — peers are already frozen in B-1). The only remaining assumption is the same single owner key as A-2; no new trust assumption is added.

See the record →

Whether an independent node can be stood up — and the public record reconstructed — using only infrastructure outside the team’s control (an independent VPS) and the public steps. Confirmed in source that all five dependency tests (private DB / API / indexer / config / default FE) are non-dependent (own PostgreSQL & Elasticsearch, records rebuilt from chain, FE is same-origin, RPC self-supplied). Reconstruction completes with the public install.sh + public repo, demonstrated by node2’s clean rebuild + the C-3 match. The one remaining dependency is “RPC access to the public chain” (public infrastructure, not the team). Reproduction by a third-party operator is an open door. (§19 #12)

See the record →

The review’s decisive proof. Two independent nodes on separate VPSs (node2 rebuilt from an empty DB by wiping the OS → reinstalling Ubuntu → re-running the installer) ingest the same chain (Base Sepolia, chainId 84532), and the reconciliation hash over the common range of 309,672 actions matches exactly (key 993db07… / content-inclusive 82651606…). The reconstruction + hashing code is confirmed to have zero source diff between the two commits. The initial raw parentHash mismatch was only a startBlock difference (isolated on the common range). Note both nodes share the same operator, so reproduction by a third-party operator remains an honest open door. (§19 #13–15)

See the record →

Whether the three downstream layers (search index, API, FE) stand independently without relying on the team’s private services. Confirmed in source that all three are non-dependent: the API serves records that its own node rebuilt from chain (even prices come on-chain from Uniswap V2); the index is built by its own Elasticsearch from the local DB (from RawEvent) = a projection of the DB whose reproducibility C-3 showed; the FE builds from the public repo with vite and points at its own node. The only outbound calls are optional features (GIF / AI images, etc.) hitting neutral third-party APIs with each user’s own key. (§19 #19–21)

See the record →

Whether the validator layer is decentralized. #16 permissioning = permissionless (none of createClient / registerInstance / processActions has an approval gate; my instanceId 232 is a live example of self-registration on the production registry [owner = own-node validator], and simulating a registration from a random address returns a fresh id [475] = permissionless still shown live). #17 censorship-resistance = yes (signed actions can be submitted by anyone, including the signer; security rests on EIP-712 signatures, made redundant by peer mirroring), verified in source. #18 queue reconstruction = partial: the finalized order is reconstructible on-chain (C-3). Strictly rebuilding the pending queue from public inputs alone is impossible in principle, but this is intentional design (the API returns only your own senderId after a privacy-audit fix / no global pending mempool / the FE fan-outs submission across multiple peers) = not a centralization hole. So #18 is categorized “By design” (not marked green) rather than left in-progress, with the reason stated. (§19 #16–18)

See the record →