Skip to content

Open-source platform for 3D AI agents. Turn text or a photo into a rigged, animated GLB avatar, give it an LLM brain, memory and a wallet, and embed it anywhere with one web component. Remote MCP server, x402 payments, Three.js, browser-native.

36.8 KB
Raw Blame

ERC-8004 Contract Deployments

All three registries are deployed via CREATE2, giving the same address on every supported EVM chain within each environment class (mainnet vs. testnet).

Provenance

Deployment transaction hashes from the original CREATE2 broadcast were not captured per chain at deploy time. Rather than leave the record as bare TODOs, each live contract is verified by on-chain bytecode: scripts/verify-onchain-parity.mjs performs an eth_getCode sweep against the public RPCs in api/_lib/erc8004-chains.js and asserts non-empty runtime code at every declared address. The bytecode ✓ (YYYY-MM-DD) notes below are the dates of those reads. Re-run any time with:

VERIFY_ONCHAIN_CHAINS=all npm run verify:onchain   # full sweep
npm run verify:onchain                              # build-gate subset (Base + Base Sepolia)

The IdentityRegistry + ReputationRegistry share one address per network class (CREATE2-deterministic). A single explorer link below resolves to the contract on any listed chain; the per-chain bytecode-verified status is the authoritative liveness signal.

Mainnet

Chains with confirmed bytecode (verified 2026-06-19): Ethereum (1), Optimism (10), BSC (56), Gnosis (100), Polygon (137), Mantle (5000), Base (8453), Arbitrum One (42161), Celo (42220), Avalanche (43114), Linea (59144), Scroll (534352).

Not yet deployed on Fantom (250), zkSync Era (324), Moonbeam (1284). An eth_getCode sweep on 2026-06-19 returned 0x (no code) at the registry addresses on these three chains across multiple independent RPCs. zkSync Era derives CREATE2 addresses differently from the EVM, so the canonical address is not reachable there without a chain-specific deploy. Registration / reputation writes on these chains will revert until the registries are deployed (deploy backlog). The address columns are kept for parity; liveness is per the bytecode notes only.

Contract Address Provenance
IdentityRegistry 0x8004A169FB4a3325136EB29fA0ceB6D2e539a432 bytecode ✓ (2026-06-19) on 12 chains; tx unrecoverable
ReputationRegistry 0x8004BAa17C55a88189AE136b182e5fdA19dE9b63 bytecode ✓ (2026-06-19) on 12 chains; tx unrecoverable
ValidationRegistry not deployed on mainnet pending mainnet deploy (src/erc8004/abi.js MAINNET.validationRegistry = '')

Testnet

Chains (all bytecode-verified 2026-06-19): BSC Testnet (97), Ethereum Sepolia (11155111), Base Sepolia (84532), Arbitrum Sepolia (421614), Optimism Sepolia (11155420), Polygon Amoy (80002), Avalanche Fuji (43113).

Contract Address Provenance
IdentityRegistry 0x8004A818BFB912233c491871b3d84c89A494BD9e bytecode ✓ (2026-06-19) on 7 testnets; tx unrecoverable
ReputationRegistry 0x8004B663056A597Dffe9eCcC1965A193B7388713 bytecode ✓ (2026-06-19) on 7 testnets; tx unrecoverable
ValidationRegistry 0x8004Cb1BF31DAf7788923b405b754f57acEB4272 reference ValidationRegistryUpgradeable (ERC-1967 proxy), interface ✓ 2026-08-06 via npm run verify:erc8004-validation; NOT this repo's ValidationRegistry.sol

Platform validator (ValidationRegistry attestor)

The platform validator is the EVM key that signs glTF/schema validation attestations (validationResponse) when an agent is registered. The deployed registry has no allowlist, so there is nothing to add it to: it can answer any request addressed to it. It needs gas on every chain it attests on, and it must be stored as the VALIDATOR_PRIVATE_KEY env var on the three-ws-api Cloud Run service (never committed). Provision/rotate with scripts/erc8004/provision-validator-key.mjs. Flow and operating notes: docs/erc8004/validation-attestation.md.

Address State Notes
0x93Bc7EfB0059B784465619FC73C2db8D01b1CD04 not configured, unfunded Provisioned 2026-06-15. As of 2026-08-06 the key is absent from .env, .env.local, Secret Manager and the Cloud Run service, and the address holds 0 gas on Base Sepolia, so /api/erc8004/validate returns validator_key_not_configured. Configure + fund it (testnet first: Base Sepolia 84532) to activate platform attestations.

CREATE2 Factory (ThreeWSFactory)

Custom vanity-prefixed CREATE2 deployer used to obtain matching addresses across chains.

Chain Address Deployer EOA Deployed Tx
BSC (56) 0x00000000D49195AE81759cd247cFeDD9D0B479df 0x4022de2D...C0564f402 2026-05-11 —
Base (8453) 0x00000000D49195AE81759cd247cFeDD9D0B479df 0x4022de2D...C0564f402 2026-05-14 0x20bbd8a8…
Arbitrum (42161) 0x00000000D49195AE81759cd247cFeDD9D0B479df 0x4022de2D...C0564f402 2026-05-14 0xa91d7cb7…

Bytecode SHA-256 424e78aad2b19a37… (1278 bytes) is identical on all three chains.

Vanity salt (deployed via Arachnid proxy 0x4e59b44847b379578588920cA78FbF26c0B4956C, prefix 0000000):

0xfc1ecd1953bb17cf798c1eaeed287873008f3a3038f438e9e74c3b33ce370ef5
  • Factory init code hash: 0x30f9d9020bf9622bbe7f8a1625d447efe350dfafd0a91e6dbd62d56547db835f

  • Grind: 96,448,706 attempts in 101.1 s, generated 2026-05-10T11:40:46Z (lucked into 8 zeros while targeting 7)

  • Source: ThreeWSFactory.sol, solc v0.8.35, optimizer 200 runs, MIT, verified on BscScan.

  • ABI:

    • deploy(bytes32 salt, bytes initCode) → address — wraps CREATE2(0, initCode, salt), reverts "create2 failed" on zero address.
    • predict(bytes32 salt, bytes32 initCodeHash) → address (view).
    • Event: Deployed(address indexed addr, bytes32 indexed salt).
  • Vanity 8-byte zero prefix (0x00000000…) saves calldata gas on every deploy/predict call.

  • To replicate on a new chain, use the same EOA + nonce + init code so CREATE2 yields the same address.

ThreeWSPayments (x402 pay-per-call receiver)

Deployed via ThreeWSFactory.deploy(salt, initCode). Constructor takes the chain's canonical USDC token, so each chain's init code differs → the cross-chain CREATE2 address parity that the factory itself enjoys does not apply here. The vanity 8-zero prefix only landed on BSC; Base and Arbitrum produced ordinary addresses from the same salt.

Owner: 0x4022de2d36c334e73c7a108805cea11c0564f402 (deployer EOA)

Vanity salt (BSC-targeted, prefix 00000000, case-insensitive):

0x5ef7540f7c609d04ab6d3997bc8c38f0f31ce09acccff2c11bcb3909ad542cde
  • Factory / deployer: 0x00000000d49195ae81759cd247cfedd9d0b479df
  • BSC init code hash: 0xb55479df540c0e4efae39a0181051754cc236a9934f03805a743f4290178569e
  • Grind: 2,859,887,864 attempts in 22.3 s, generated 2026-05-10T13:58:54Z
Chain USDC ThreeWSPayments Tx
BSC (56) 0x8AC76a51cc950d9822D68b83fE1Ad97B32Cd580d 0x00000000381f09742a30a5a49975514AeC1B72Cc 0xc4f4e87f…
Arbitrum (42161) 0xaf88d065e77c8cC2239327C5EDb3A432268e5831 0xed3696489490dbfAFD82996ADB11165A56c33c49 0xca39a600…
Base (8453) 0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913 0x31B13cDe47431EfcC8616C8495204e6E6C2Ded34 0xb6fcf60b…

Base deploy status — LIVE, bytecode confirmed (re-verified 2026-06-15)

The earlier "tx mined but address has no code → treat as not deployed" note was wrong — a stale/unsynced RPC reading at deploy time. The deploy succeeded. Proof, re-checked on-chain against live Base RPC:

  • Deploy tx 0xb6fcf60b… emitted Deployed(addr=0x31B13cDe…, salt=0x5ef7540f…) from ThreeWSFactory — so the predicted address is the deployed address (no CREATE2 collision, no salt mismatch).
  • eth_getCode(0x31B13cDe…) returns 1243 bytes (non-empty). owner() → 0x4022de2D36C334E73C7a108805Cea11C0564f402 (the deployer EOA, as on BSC/Arbitrum), and Base USDC 0x833589fCD6… is embedded as the USDC immutable (BSC USDC is not).
  • Independently re-derived: compiling the in-repo source (contracts/ThreeWSPayments.sol, solc 0.8.35, optimizer 200) reproduces the exact recorded BSC init-code hash 0xb55479df… and live BSC address — confirming source + settings are byte-for-byte correct — and CREATE2(factory, salt, Base-USDC init code) → 0x31B13cDe…. The Base init-code hash is 0x253291817df177b537145a05d0221065be924cffa606b387221b5c6bf8f1c475. (1243 bytes runtime, vs BSC's 1278, differs only because each chain's USDC enters the constructor as an immutable.)

Basescan verification: PENDING (not yet verified). Source + exact settings are captured in-repo; run scripts/verify-threews-payments-base.mjs with a BASESCAN_API_KEY set to publish (constructor args: (0x4022de2D…, 0x833589fCD6…)). The script prints the Standard JSON Input bundle for manual UI verification when no key is present.

Base payment routing — EOA payTo is intentional (no redeploy / no redirect)

X402_PAY_TO_BASE stays the EOA 0x4022de2D36C334E73C7a108805Cea11C0564f402, not the contract. Base x402 settles via the exact scheme (EIP-3009 transferWithAuthorization / Permit2) through the CDP facilitator, which performs a plain ERC-20 USDC transfer to payTo — it never calls pay(bytes32) on the recipient. An EOA receiver is correct there: funds land directly and stay liquid. Pointing payTo at the contract would route USDC in via a raw transfer with no Payment event, recoverable only through the contract's withdraw() — strictly worse.

ThreeWSPayments is load-bearing only on BSC (X402_PAY_TO_BSC = the contract), where Binance-Peg USDC implements no EIP-3009 and no facilitator advertises eip155:56, so the contract-mediated pay(bytes32) "direct" scheme is the only option (see api/_lib/x402-bsc-direct.js). The Base instance exists for cross-chain parity and as an on-chain record; the Base x402 flow has no code path that calls it, and Base payments have been settling correctly to the EOA all along.

Deploy command (run from 3D-Agent repo where scripts/deploy-multichain.mjs lives):

PK=<deployer-private-key> \
PAYMENTS_SALT=0x5ef7540f7c609d04ab6d3997bc8c38f0f31ce09acccff2c11bcb3909ad542cde \
BASE_RPC_URL=... ARB_RPC_URL=... BSC_RPC_URL=... \
node scripts/deploy-multichain.mjs

AgentPayments (EVM agent-token payment engine)

EVM port of the Solana pump_agent_payments program. One deployment per chain serves every agent token on that chain. Deploy + wiring guide: AGENT_PAYMENTS.md. Source: src/AgentPayments.sol. Owner = protocol/global buyback authority (set to the platform multisig).

Unlike ThreeWSPayments, the constructor takes only owner (no chain-specific immutable), so a CREATE2 deploy via ThreeWSFactory with a shared salt yields the same address on every chain. Plain new deploys produce per-chain addresses (nonce-dependent) — fill in whichever you used.

Chain Chain ID AgentPayments Owner Routers allow-listed Tx
Ethereum 1 TODO: fill after deployment TODO TODO TODO
Base 8453 TODO: fill after deployment TODO TODO TODO
Arbitrum One 42161 TODO: fill after deployment TODO TODO TODO
Polygon 137 TODO: fill after deployment TODO TODO TODO
BNB Smart Chain 56 TODO: fill after deployment TODO TODO TODO
Avalanche 43114 TODO: fill after deployment TODO TODO TODO

After filling a row, set the matching agentPayments address in agent-payments-sdk/src/evm/addresses.ts and run npm run build in agent-payments-sdk/.

WorldMoves (event-only move-commit contract)

BNB Chain campaign, Track C: an event-only on-chain move stream for three.ws real-time worlds, designed to be called every ~0.45s (BSC's live block time). Source: src/WorldMoves.sol. No admin, no owner, no upgradeability — move() never writes storage; checkpoint() is the only opt-in storage-writing call. Full design rationale in the contract NatSpec.

Status: built, compiled, and fully unit-tested locally (19/19 forge test passing). Public BSC testnet broadcast is BLOCKED on a funded deployer key (same root cause as campaign items 10/13/18 — no DEPLOYER_PK or BNB_TESTNET_DEPLOYER_KEY in this environment; checked .env, contracts/.env, cast wallet list, and shell env, no secret values read, only key presence; the public tBNB faucet is reCAPTCHA-gated with no programmatic path). A forge script ... -vvvv dry-run against the LIVE BSC testnet RPC (https://data-seed-prebsc-1-s1.bnbchain.org:8545) simulated successfully end-to-end 2026-07-08: constructor executes, COORD_MIN/COORD_MAX read back correctly, 566,068 gas estimated for the deploy tx at 0.1 gwei ≈ 0.0000566068 BNB — the RPC, script, and bytecode are all deploy-ready.

Re-verified 2026-08-02. The dry run above still simulates green against the same live testnet RPC (chainId 97, block 122,607,654 at read time) at exactly 566,068 gas, and forge test is still 19/19. The funding blocker is unchanged and was re-checked from scratch, not carried over: shell env, .env, .env.local, the three-ws-api Cloud Run service env, and Secret Manager all hold no funded chain-97 key, and three programmatic faucet endpoints (bnbchain's API, Stakely, Triangle) refuse or are retired. A throwaway deployer was written to the gitignored contracts/.env at 0xC4e63FdF188D94059C877b957866726A888e1240; that key no longer exists on this machine (see the 2026-09-02 note below for the address that replaced it, and do not fund the retired one). scripts/bnb-testnet-deploy-prove.mjs is the single command for the rest: a signature-free preflight by default, and with --broadcast it deploys both contracts and then proves the live sender, reader, and ghost paths against what it just deployed.

Re-verified 2026-09-02, and the deployer address changed. Everything above was re-measured against the current tree rather than carried forward:

  • Both dry runs simulate green against the live BSC testnet RPC at exactly the same gas as before: DeployWorldMoves 566,068 gas and DeployGreenfieldVault 1,711,362 gas, both at 0.1 gwei (0.0000566068 and 0.0001711362 BNB).
  • forge test is 19/19 on WorldMoves and 41/41 on GreenfieldVault (the vault suite has grown by 7 cases since the 2026-08-02 note's 34).
  • All four BSC testnet RPC lanes in api/_lib/bnb/chains.js answer chainId 0x61 today, so the lane list is no longer one rung thin.
  • The deployer is now 0x1C4918894dfA5eE11cfF9629B458b5169Cfa3871. The 2026-08-02 throwaway key was not preserved across sessions, and its address holds 0 tBNB, so nothing was stranded by replacing it. The new key is testnet-only, lives in the gitignored contracts/.env at mode 600, and has never held and must never hold mainnet value.
  • The faucet still has no agent path: testnet.bnbchain.org/faucet-smart serves its reCAPTCHA page rather than a claim API, and the Stakely endpoint is gone. Funding remains one human action.
  • The full --broadcast path was re-validated end to end against a local anvil --chain-id 97 before asking for funds, so a funded run will not be the first time this code executes: both contracts deployed, then 1 join, 3 move, and 1 leave mined through the real api/_lib/bnb/world-moves.js sender, decoded live by src/bnb/world-presence-reader.js (1 Joined, 3 Moved, 1 Left, zero reader errors), with createGhostTracker holding 1 ghost interpolating {x:1110, z:-445} toward its {x:1500, z:-250} target and dropping to 0 on Left. Local addresses are deliberately not recorded here; they are anvil-local and mean nothing on the public chain.

Re-verified 2026-09-04. Still one faucet claim away, nothing else. Every line was re-measured against the current tree, not carried forward:

  • Both dry runs simulate green against the live chain-97 RPC at exactly the same gas as the two previous re-reads: DeployWorldMoves 566,068 gas and DeployGreenfieldVault 1,711,362 gas at 0.1 gwei (0.0000566068 and 0.0001711362 BNB, 0.000227743 BNB for the pair).
  • forge test is 19/19 on WorldMoves and 41/41 on GreenfieldVault.
  • The deployer 0x1C4918894dfA5eE11cfF9629B458b5169Cfa3871 still holds 0 tBNB, read independently on three RPC lanes (data-seed-prebsc-1-s1, bsc-testnet.drpc.org, bsc-testnet-rpc.publicnode.com), all answering chainId 97.
  • The --broadcast path was re-proven end to end against a local anvil --chain-id 97, running the real scripts/bnb-testnet-deploy-prove.mjs --broadcast rather than a hand-assembled sequence: both contracts deployed, then 1 join, 3 move, and 1 leave mined through the real api/_lib/bnb/world-moves.js sender, decoded live by src/bnb/world-presence-reader.js (1 Joined, 3 Moved, 1 Left), with createGhostTracker holding 1 ghost before the leave and 0 after. Local addresses are again deliberately not recorded: they are anvil-local.
  • The faucet is still the only human step. Four claim paths were checked and every one is captcha- or account-gated: the official testnet.bnbchain.org/faucet-smart (its /api/v1/faucet path serves the HTML page, not a claim API), ghostchain.io and faucet.zalalena.com (both Cloudflare Turnstile / hCaptcha), and tokentool.bitbond.com (wallet connect plus a completed third-party profile).
  • The wiring after funding is unblocked and was re-checked: gcloud is authenticated in the workspace again, WORLD_MOVES_ADDRESS_TESTNET is confirmed absent from the three-ws-api service env, and /api/bnb/world-config?network=testnet answers live today with address: null, deployed: false. Setting the var is a config-only --update-env-vars call, which needs no further approval.

Re-verified 2026-09-09. The faucet has a precondition the earlier notes missed, and it is the reason funding has never landed. Everything was re-measured against the current tree:

  • Both dry runs simulate green against the live chain-97 RPC at the same gas as all three previous re-reads: DeployWorldMoves 566,068 gas and DeployGreenfieldVault 1,711,362 gas at 0.1 gwei (0.0000566068 and 0.0001711362 BNB, 0.000227743 BNB for the pair). Run without --rpc-url the scripts revert with "no known Greenfield hub addresses for this chain id", which is correct behavior on foundry's default chain 31337, not a regression: the hub table is keyed by chain id, so a dry run has to name a chain-97 RPC.
  • forge test is 19/19 on WorldMoves and 41/41 on GreenfieldVault.
  • scripts/bnb-testnet-deploy-prove.mjs (no flags) still reports the honest preflight and signs nothing: DEPLOYER IS UNFUNDED. Nothing was signed.
  • The deployer 0x1C4918894dfA5eE11cfF9629B458b5169Cfa3871 holds 0 tBNB on chain 97 (read on bsc-testnet-rpc.publicnode.com and data-seed-prebsc-1-s1) and nonce 0, so it has never sent a transaction.
  • New finding: the official faucet gates on a BSC MAINNET balance. bnbchain.org/en/testnet-faucet requires the claiming address to hold 0.002 BNB on BSC mainnet, dispenses 0.3 tBNB, and allows one claim per 24 hours. The deployer holds 0 BNB on mainnet (read on bsc-rpc.publicnode.com), so completing the reCAPTCHA against this address would fail eligibility even with a human at the keyboard. That is why three sessions of "one human faucet claim away" never converted.
  • The cheaper route avoids a mainnet transfer entirely: claim the 0.3 tBNB from any address that already holds 0.002+ BNB on mainnet, then send tBNB to the deployer above. Only 0.000227743 tBNB is needed for both deploys, so a single 0.3 tBNB claim covers this deploy and every future one by three orders of magnitude.
  • Wiring after funding is still unblocked: WORLD_MOVES_ADDRESS_TESTNET is confirmed absent from the three-ws-api service env, and /api/bnb/world-config?network=testnet answers live today with address: null, deployed: false.

Real broadcast proof — anvil fork of LIVE BSC testnet state (per 00-CONTEXT's decision-default table: "if every faucet fails, finish ALL code + tests against a local anvil --chain-id 97 fork" — the same workaround prompts 02/03 used successfully). anvil --chain-id 97 --fork-url https://bsc-testnet.drpc.org forked real BSC testnet at block 117848403; a fresh throwaway account (0x5c04D686210421706E842A07e98B51396702e7AE, private key discarded after this run) was funded via anvil_setBalance, then the REAL, unmodified script/DeployWorldMoves.s.sol was run with --broadcast against the fork — same script, same bytecode that would run against the public RPC, only the RPC endpoint differs:

$ forge script script/DeployWorldMoves.s.sol:DeployWorldMoves \
    --rpc-url http://127.0.0.1:8555 --private-key $THROWAWAY_PK --broadcast -vvvv
[353991] → new WorldMoves@0x71Ddcb9865632Ca3c4325dE0E4a92Cc0065c8aaE
ONCHAIN EXECUTION COMPLETE & SUCCESSFUL.
  • Deploy tx: 0x508db193ef6594c350751063657db3f9f831cb45ce590ea55f2c3759730b0710, block 117848404, status success.

  • Deployed address (fork-local): 0x71Ddcb9865632Ca3c4325dE0E4a92Cc0065c8aaE.

  • 10 real move() transactions fired back-to-back via cast send through the real deployed contract (not a reimplementation), each mined into its own block, receipts fetched independently and one full log decoded to confirm the exact Moved event shape:

    # tx hash block timestamp gasUsed status
    1 0xae6a706c…4a0 117848405 1783475882 26293 success
    2 0xd557c5d9…d9 117848406 1783475883 26293 success
    3 0x111b513d…a7 117848407 1783475884 26293 success
    4 0x9985f6e2…f8 117848408 1783475884 26293 success
    5 0x808f4586…f2 117848409 1783475885 26293 success
    6 0x39203f83…5d 117848410 1783475885 26293 success
    7 0x3a58b91b…d7 117848411 1783475885 26293 success
    8 0x878e042f…e8 117848412 1783475886 26305 success
    9 0xe7e9c934…be 117848413 1783475887 26305 success
    10 0x0b66bc4b…42 117848414 1783475888 26305 success

    One block minted per transaction (matches move()'s design goal of supporting one call per block); gasUsed is the full transaction-level cost (21,000 base intrinsic gas + calldata + 3-topic log), consistent with the forge test unit measurement below of ~4,800 gas of internal execution alone. Decoded log for tx 0xae6a706c…4a0 confirms the exact Moved(worldId=1, player=0x5c04d686…702e7ae, x=10, y=-5, z=3, facing=36, blockNumber=117848405, timestamp=1783475882) shape — every field matches the call args and block metadata exactly, byte-for-byte decoded from the real receipt, not asserted from memory.

    Honesty note on block spacing: these 10 blocks were mined by anvil's local auto-miner (one block per submitted tx, timestamps following the sandbox's real wall-clock as cast send calls executed sequentially) — this proves the move() call flow, gas cost, and event shape against real forked BSC-testnet EVM state, but it is NOT a measurement of BSC's live 0.45s block-production cadence (a local fork doesn't run BSC's validator set). The 0.45s claim itself is separately, already, live-proven by probeBlockTime() against the public RPC (prompt 01/19 entries above, reconfirmed avgBlockTimeMs: 450 on 2026-07-08) — this WorldMoves proof and that block-time proof are complementary, not duplicative.

Deploy (once funded, real public broadcast):

forge script script/DeployWorldMoves.s.sol:DeployWorldMoves \
  --rpc-url https://data-seed-prebsc-1-s1.bnbchain.org:8545 \
  --private-key $DEPLOYER_PK \
  --broadcast

Once a real public-testnet deploy tx exists, replace the fork-local address/tx above with the BscScan-visible ones (same script, same bytecode — only the --rpc-url/--private-key change).

Prompt 18 — two-wallet world E2E proof (anvil, --block-time 0.45)

Prompt 18 (bnb-chain campaign, work order 18, world e2e demo; retired, see git history) needed two real participants moving in the same world at once with block spacing that actually reproduces BSC's live ~0.45s cadence — a step up from 14/15/16's wall-clock-paced local blocks. This run started a FRESH (non-forked) anvil --chain-id 97 --block-time 0.45 --port 8555 — anvil's interval-mining flag accepts fractional seconds, so this genuinely interval-mines a block every 0.45s regardless of pending txs, the same rate 00-CONTEXT's probeBlockTime() measured live on the public RPC (avgBlockTimeMs: 450), not merely "close to it."

  • Deploy: the real, unmodified DeployWorldMoves.s.sol script, --broadcast, deployer = anvil's well-known account #0 (public test key, zero real funds). Deploy tx 0xab86fb5937e655dd1e64d8e45118ca5624d20f9b4bd213704067f5dc7ae8b65a, block 18, status success, contract 0x5FbDB2315678afecb367f032d93F642f64180aa3.

  • Two independent real Chromium browser contexts (Playwright, headless), each pre-seeded with a distinct funded anvil test-mnemonic account (0x7099…dc79C8 / 0x3C44…4293BC) as its three.ws:bnb-presence-key localStorage session key, navigated to the real /agora?play=1 route with the existing ?bnbDevAddress=&bnbDevRpc= pre-public-deploy override (src/agora/onchain-presence.js's turnOn()), then driven with real WASD keyboard input for 60s each, concurrently.

  • 7 real move() txs landed on-chain, all status: success, mode: self-pay (MegaFuel testnet declined sponsorship — no policy provisioned for these throwaway addresses, the expected/documented outcome, see prompt 15/16 entries):

    wallet tx hash block gasUsed block timestamp (unix)
    A 0xd1b5188c…9d7 1849 26281 1783496844
    A 0xb63fdb5b…997 1880 26305 1783496858
    A 0x5f772762…f61 1927 26305 1783496879
    A 0x11dd19f4…b32 1986 26305 1783496905
    B 0x945ce448…715 1863 26281 1783496850
    B 0x317554f6…b20 1916 26305 1783496874
    B 0x3d378b15…c78 1963 26305 1783496895

    Real block-cadence check across the run: block 1849 (ts 1783496844) → block 1986 (ts 1783496905) is 137 blocks over 61 real seconds = 0.445s average inter-block time, matching the live BSC testnet 0.45s Fermi cadence to within 1%. Finer-grained per-block sampling (cast block 1849 1850 1851 1852 1853 1854 1855) shows two blocks per integer second (1849,1850→...844, 1851,1852→...845, 1853,1854,1855→...846) — consistent with true sub-second (0.45–0.5s) interval mining, not per-tx-triggered mining.

  • Cross-wallet visibility confirmed live, independently, both directions: session A's console logged ghost joined player=0x3C44…4293BC (wallet B) and session B's console logged ghost joined player=0x7099…dc79C8 (wallet A) — each browser picked up the other's real on-chain Moved/Joined events via watchWorldPresence(), with zero coordination between the two Playwright contexts beyond the shared chain.

Honesty note: identical in spirit to prompt 14/15's caveat — this is a local anvil instance, not the live BSC testnet validator set, so it does not substitute for probeBlockTime()'s live measurement. What's new here is that the local cadence was explicitly configured to match the live-measured rate (--block-time 0.45) rather than following the sandbox's incidental wall-clock pacing, so the block-spacing numbers above are a deliberate reproduction of the real rate, not a coincidence of how fast cast send was invoked. Full run log, receipts, and the two-wallet Playwright script (deleted after the run per repo hygiene — reproducible from this write-up) recorded in prompts/bnb-chain/PROGRESS.md's prompt 18 entry.

GreenfieldVault (pay → PermissionHub grant, on-chain-gated 3D asset vault)

BNB Chain campaign, Track B: a BSC marketplace contract that, on payment, calls the REAL Greenfield PermissionHub.createPolicy cross-chain, granting the buyer read permission on an encrypted object. Source: src/GreenfieldVault.sol; real interface stubs (sourced from bnb-chain/greenfield-contracts, verified against master on 2026-07-08 — no invented ABI) in src/greenfield/. Uses the real, already-deployed PermissionHub/CrossChain/ObjectHub hubs — no Greenfield contract of our own to deploy.

Constructor-supplied hub addresses (deploy script script/DeployGreenfieldVault.s.sol defaults per block.chainid, env-overridable):

Network PermissionHub CrossChain ObjectHub (IGnfdAccessControl)
BSC mainnet (56) 0xe1776006dBE9B60d9eA38C0dDb80b41f2657acE8 0x77e719b714be09F70D484AB81F70D02B0E182f7d 0x634eB9c438b8378bbdd8D0e10970Ec88db0b4d0f
BSC testnet (97) 0x25E1eeDb5CaBf288210B132321FBB2d90b4174ad 0xa5B2c9194131A4E0BFaCbF9E5D6722c873159cb7 0x1b059D8481dEe299713F18601fB539D066553e39

(Mainnet CrossChain/ObjectHub bytecode-verified 2026-07-07 per prompts/bnb-chain/00-CONTEXT.md; PermissionHub mainnet + all testnet addresses read live from bnb-chain/greenfield-contracts's README "Contract Entrypoint" tables on 2026-07-08 — never invented.)

Status: built, compiled, and fully unit-tested locally (34/34 forge test passing, including two dedicated re-entrancy proofs against a mocked PermissionHub/CrossChain/IGnfdAccessControl). NOT deployed to BSC testnet. A forge script ... -vvvv dry-run against the live BSC testnet RPC (https://data-seed-prebsc-1-s1.bnbchain.org:8545, also reachable via the new bsc_testnet foundry.toml alias) simulated successfully end-to-end (constructor executes against the real testnet PermissionHub/CrossChain/ ObjectHub addresses above, ~1.16M gas estimated for the deploy tx at 0.1 gwei ≈ 0.000170 BNB) — the RPC, script, and bytecode are all deploy-ready. The only missing piece is a funded deployer key: neither DEPLOYER_PK nor BNB_TESTNET_DEPLOYER_KEY is present in this environment (checked .env, contracts/.env, cast wallet list, and shell env; no secret values were read, only key presence). Same funding blocker as campaign items 13 and 18 — owner-only to unblock (fund a deployer EOA via the tBNB faucet, then run the broadcast command below).

Re-verified 2026-08-02. The dry run still simulates green against the live testnet RPC at 1,695,618 total gas (1,226,903 of it the constructor) at 0.1 gwei = 0.0001695618 BNB, and forge test is still 34/34. The 1.16M/0.000170 BNB figures above were the constructor-only measurement; the number that matters for funding is the total. The funding blocker is unchanged; see the re-verification notes in the WorldMoves section for the full re-check and for the throwaway deployer address now waiting on the faucet. Re-measured again 2026-09-02 and 2026-09-04: same 1,711,362 total gas both times, and forge test is 41/41. See the WorldMoves section's 2026-09-04 block for the full re-read, including the four faucet paths that are all human-gated.

Deploy (once funded), preferred, because it also proves the live paths in the same run:

node scripts/bnb-testnet-deploy-prove.mjs            # preflight, signs nothing
node scripts/bnb-testnet-deploy-prove.mjs --broadcast # deploy both + live proof

The underlying forge command it runs, if you want it by hand:

forge script script/DeployGreenfieldVault.s.sol:DeployGreenfieldVault \
  --rpc-url https://data-seed-prebsc-1-s1.bnbchain.org:8545 \
  --private-key $BNB_TESTNET_DEPLOYER_KEY \
  --broadcast

After a real deploy: replace this status block with the address, deploy tx hash, and a BscScan link; have a seller grantRole(ROLE_CREATE, vault, expiry) on the real testnet ObjectHub for a real uploaded object (prompt 09); execute one real buy(); and paste the BSC tx hash plus the resulting PermissionHub.settleCreatePolicy/PolicyGranted proof (BscScan + GreenfieldScan once the cross-chain ack settles) per the bnb-chain campaign work order 10 (vault contract; retired, see git history) definition of done.

Prompt 11 (vault unlock API) local-anvil proof, 2026-07-08 — mocked-hubs deploy script added: script/DeployGreenfieldVaultMocked.s.sol deploys MockCrossChain/MockPermissionHub/MockGnfdAccessControl (the same mocks test/GreenfieldVault.t.sol's 34-test suite already validates) plus a real GreenfieldVault wired to them, on a local anvil chain — LOCAL TEST-ONLY, never point it at a real network. Used to prove api/vault/* (list/status/unlock) against a genuinely deployed contract + real transactions before a funded deployer key exists; full run log in prompts/bnb-chain/PROGRESS.md's prompt 11 entry. One real bug was caught and fixed by this run: getVaultLogs() resolved its default toBlock via viem's getBlockNumber(), which memoizes for several seconds by default — two calls made back-to-back (e.g. right after a buy()/settlement tx, the exact shape of a real poll) could read a stale toBlock and silently miss the newest events. Fixed with cacheTime: 0 on that one call.

Prompt 13 (vault-e2e-proof) capstone chain run, 2026-07-08 — same mocked-hubs script, full list→buy→settle→unlock→decrypt chain proven together: re-deployed the identical DeployGreenfieldVaultMocked.s.sol (same deterministic addresses: GreenfieldVault 0xCf7Ed3AccA5a467e9e704C703E8D87F634fB0Fc9, MockPermissionHub 0xe7f1725E7734CE288F8367e1Bb143E90bb3F0512) against a fresh anvil chain, then ran a real GLB (from the free forge lane) through encryptGlb → list() → buy() → settleCreatePolicy() → the REAL exported api/vault/{list,status,unlock,download}.js handlers → client-side unwrapKey/decryptGlb → a byte-identical sha256 match against the original GLB. Full proof (every tx hash, every HTTP response, the real "Query failed with (6): No such bucket" finding from testing download.js against live Greenfield testnet infra with a throwaway operator key) is in prompts/bnb-chain/PROGRESS.md's prompt 13 entry. One real bug caught and fixed by this run: settleCreatePolicy(saleId, status)'s second argument is STATUS_SUCCESS = 0 (not 1) per GreenfieldVault.sol's own constant — passing 1 silently drives the contract's Failed branch and clears saleIdOf, with waitForTransactionReceipt still resolving normally (a successful receipt is not the same as the contract's OWN state machine landing where the caller assumed).

Notes

  • Addresses are authoritative in src/erc8004/abi.js (REGISTRY_DEPLOYMENTS).
  • Changing any address requires redeployment and updating REGISTRY_DEPLOYMENTS in abi.js and api/_lib/erc8004-chains.js.
  • Deploy scripts: script/Deploy.s.sol (testnet), script/DeployValidationMainnet.s.sol (mainnet ValidationRegistry).
  • 15-chain deploy command list: script/deploy-validation-registry.sh.
  • After deployment: run computeAddress(DEPLOYER_ADDRESS) in the script (dry-run) to confirm the address, then update validationRegistry in src/erc8004/abi.js and sdk/src/erc8004/abi.js.