1. Question restated

Is Agent First Opening worth building and launching within 14 days on IMD or CLAUS, and, if it is, what is the smallest version that proves it?

2. Verdict in five lines

1. Gap: partly served. Record (read 2026-10-23): launch products already offer decaying sniper taxes, wallet caps and uniform-price auctions; none of the reviewed live products proves one allocation per independent agent identity. Pons v2, Flaunch API, Uniswap CCA.
2. Speed: 14 days for a falsification MVP on IMD Sepolia; no public launch in that window. Inference (2026-10-23): the current evidence does not clear identity farming, router coverage or production hook admission. CLAUS cannot replay its opening and its extension owner has no timelock. CLAUS host verified source, Uniswap hook routing.
3. MVP: Inference (2026-10-23): an immutable IMD Sepolia v4 hook admits signed, preapproved seat-owner or agent-wallet EOA transactions for the first 120 blocks, caps the actual token output per identity, then becomes an ordinary 1.25% pool; three independently owned agent reviewers post eligibility decisions and receive onchain escrow bounties after verification. No ZK in this experiment.
4. Largest failure risk: Record and Inference (read 2026-10-23): ERC-8004 registration is permissionless and IMD seats have concentrated ownership; even an origin-bound identity signature does not prove a unique human or the eventual router recipient. Smart-account and relayer flow is unresolved. ERC-8004 specification, IMD seat owners, v4 router interface.
5. Total IMD cost: Inference (2026-10-23): three MVP requests × 0.5 IMD = 1.5 IMD before gas and agent bounties: two job.open and one launch.open; a later optional oracle adds 0.5 IMD. IMD docs, capabilities.

3. Competitor table

Method (read 2026-10-23): “Last activity” below is the latest dated event or repository commit I could verify, not a claim about latest private development. A live status is a record of the cited product/deployment, not a security endorsement. GitHub dates are pinned to commits in §13.

Name	Link	Chain	Status	Last verified activity	Overlap	What it misses
Pons v2	official v2 docs	Robinhood Chain	Live; new launches whitelisted, Claim (read 2026-10-23)	Record: factory event in block 96,196,144, 2026-10-23 19:52 UTC; event purpose not decoded	Claim (read 2026-10-23): 99% buy tax at opening, exponentially to zero in 5 seconds; creator exemptions exist	Identity cap; exemptions make some bundled opening buys privileged. This is a bonding-curve opening before a v4 pool, not the proposed hook.
Flaunch	API docs, SDK	Base, Ethereum, Robinhood Chain	Live, Claim (read 2026-10-23)	Record: SDK commit f9bc6ac58d1d3c16f7899919a9bec6a2d9e81ecb, 2026-10-21	Claim: temporary 0.25%-of-supply cap per wallet	New wallets can acquire additional caps; no agent identity or onchain reviewer payment.
Doppler	explainer, SDK	Base, Ethereum, Monad	Live, Claim (read 2026-10-23)	Record: SDK commit 3eaa03e37057b304e81e0fc61a12612c3f5ecf91, 2026-10-20	Claim: dynamic auction and declining opening fees, including examples from 80% to 1%	Price/timing defense rather than one allocation per identity.
Uniswap Continuous Clearing Auction	contracts and validation hook, launch case	Ethereum, Unichain, Arbitrum, Base	Live, Record (read 2026-10-23)	Record: ce78b4d94cdf94fd41f20afb6f6e495b8b83a234, 2026-10-22	Uniform-price, continuous clearing reduces timing games; optional bid-validation hook can gate participants	It is an auction ahead of liquidity, not an ordinary v4 swap pool with post-opening fee monotonicity or paid agent review. It is the strongest simpler alternative.
Zama encrypted ICO	event account, auction docs	Ethereum	Live auction completed, Claim (read 2026-10-23)	Claim: event ran 2026-01-21 through 2026-01-24; article dated 2026-01-26	Encrypted bid sizes conceal demand quantity, while bid prices remain public	Uses encrypted auction, mandatory identity/wallet approval and a separate settlement path; no v4 opening hook or permissionless identity cap. Its claimed funding is not evidence of this idea's demand.
CLAUS	fee state, verified TradeExtensionHost	Ethereum	Live, Record (read 2026-10-23)	Record: fee-state observed 2026-10-23 at block 26,257,472; host source verified 2026-10-15	Hook sees sender and hookData; fee plan can change through a module up to 14.9%	Opening has passed; token owner can replace the module without a 24-hour timelock, and hook sender is a router.

Comparison to real openings (Inference, 2026-10-23): Pons and Doppler make the first seconds expensive, Flaunch caps a wallet, and CCA/Zama avoid first-transaction price discovery with auctions. Agent First Opening trades their simple timing/price mechanisms for an identity assumption. The assumption is the novel part and also the attack surface; use CCA if the only goal is equal-price opening, and use this hook only if a launch specifically values reputation-weighted access. Pons, Flaunch, CCA technical documentation, Zama auction documentation.

4. Hook spec

Deployment target (Record, read 2026-10-23): IMD supports univ4_hook launch requests on Ethereum and Robinhood Chain and a Sepolia test chain, accepts a customToken only when transfer behavior is ordinary and supply cannot be minted after launch, and takes 10% of supply for its swarm. Its Ethereum LaunchFactory source at verified address 0x12c63b581d07093f6126bc02263c58f7eadaa96f deploys, initializes and seeds in one transaction, requires beforeInitialize or afterInitialize, and currently uses a 1.25% pool fee for launches. IMD docs, capabilities. The candidate token uses IMD's plain fixed-supply ERC-20. No transfer fee, transfer restriction, venue lock, or post-launch mint exists.

Callbacks and flags (Inference design, 2026-10-23): beforeInitialize (1 << 13) requires the immutable, verified IMD factory as its sender, binds one pool key, validates the configured token pair/fee and stores endBlock = block.number + 120; no other caller may bind the hook first. beforeSwap (1 << 7) checks opening buy eligibility and nonce, and returns a positive specified-currency input premium as a BeforeSwapDelta; beforeSwapReturnsDelta (1 << 3) is therefore required. afterSwap (1 << 6) checks the realized token output and updates filled[identityKey]. No other callback or permission flag. The address low-bit mask is 0x20c8; mine a CREATE2 salt for that mask. A return-delta flag without its callback is invalid in v4. v4 Hooks library, IHooks interface, PoolManager accounting.

Opening condition and data (Inference design, 2026-10-23): only exact-input token buys at block.number < endBlock require nonempty hookData; sells remain open. The ABI tuple is (uint8 version, uint8 identityKind, uint256 identityId, bytes32 canonicalIdentityKey, address snapshotSigner, address submitter, uint256 maxInput, uint256 nonce, uint64 deadline, bytes32[] allowlistProof, bytes signature). The Merkle leaf is keccak256(abi.encode(originChainId, registryAddress, identityKind, identityId, canonicalIdentityKey, snapshotSigner)). An EIP-712 permit commits to chainId, hook, PoolId, sender (the calling router), submitter, those fields and the fixed opening cap; abs(amountSpecified) ≤ maxInput, deadline and unused nonce are checked before swap. The smallest MVP admits only identity-controlled EOAs: submitter == tx.origin == snapshotSigner, that EOA signs, and snapshotSigner must equal the current ownerOf(seatId) or nonzero ERC-8004 getAgentWallet(agentId) on the chain with the registry. A seat/agent-wallet transfer after the snapshot invalidates this opening proof; token transfers remain unrestricted. Binding both router and origin prevents another EOA from copying an exposed permit into its own transaction. A prelaunch immutable Merkle root marks approved keys; an ERC-8004 registration alone never qualifies. A reviewer must consolidate a seat and a linked agent ID to one canonical key. Every spent nonce and each key's cumulative realized output persist across blocks and wallets. Empty data works after opening. On Sepolia, use a root containing Ethereum seat ownership at a pinned Ethereum block; the open-mint Sepolia test seat NFT cannot establish scarce identity. Fork tests separately exercise live Ethereum registry calls. Smart-contract wallets, account-abstraction bundlers and relayers are excluded during the opening until a recipient-verifying route is proven; this is an account-path limitation, not a token transfer or venue lock. ERC-8004 EIP, official registry deployments, IMD test-chain contracts, v4 router interface.

Router truth (Record and limitation, read 2026-10-23): v4 passes PoolManager.swap its immediate caller as sender; a Universal Router or other intermediary is therefore visible instead of the trading wallet. The v4 router carries hookData per swap hop, but its swap action and later token-settlement actions can direct output independently. A hook can verify that an identity EOA submitted and authorized the purchase by checking tx.origin plus its EIP-712 signature; it cannot prove the token-recipient wallet or beneficial funder from arbitrary router calldata. tx.origin is intentionally used only for this EOA-only opening path, not as admin authority. The MVP must describe access this way. It must accept any router that forwards valid hookData for an identity EOA; nonforwarding routes revert only in the opening. Uniswap routing may additionally require hook allowlisting. Support in 0x, 1inch and other aggregators is not found. PoolManager, V4Router, Uniswap hook allowlist, routing guidance.

Fee math (Inference proposal, 2026-10-23): retain the IMD static LP fee at 12,500 parts per million (1.25%). beforeSwap takes a flat 2,500 ppm (0.25%) of gross input during the opening and zero thereafter, rounded down. PoolManager credits the hook's custom-accounting balance; a permissionless sweep can take only to the immutable fee vault. Neither rate can rise. For an eligible exact-input 1 ETH buy during opening: hook premium 1 × 0.0025 = 0.0025 ETH; the pool sees 0.9975 ETH; LP fee 0.9975 × 0.0125 = 0.01246875 ETH; pre-price-swap input 0.9975 − 0.01246875 = 0.98503125 ETH; aggregate fee 0.01496875 ETH, or 1.496875% of gross. After opening: pool fee 1 × 0.0125 = 0.0125 ETH, pre-price-swap input 0.9875 ETH. Price impact is additional, not a fee. The verified IMD LP split allocates 80% of LP fees to the requester recipient and 20% to the network. The proposed requester recipient is an immutable, opt-in staking vault: ordinary token holders may deposit/withdraw freely; its pull-based two-currency accounting sends 80% of requester LP fees and the opening premium to stakers, and at most 20% of requester LP fees repays the sponsor to a fixed cap. Premium accumulated before anyone stakes streams over a fixed day after opening rather than being captured by the first depositor. No ERC-20 transfer rule changes. The custom-accounting arithmetic, reward accounting and factory admission are untested MVP gates. v4 Hooks library, PoolManager, IMD factory, IMD docs.

Fairness and invariants (Inference design, 2026-10-23): three buys in one private bundle still share filled[key]; filled[key] + actualTokensOut ≤ cap after each swap, regardless of router or nonce. The cap is a token quantity fixed before launch, not an input quote vulnerable to price movement. afterSwap reverts the whole fill if it would exceed the cap; a later transaction can choose a smaller exact-input amount. Equal identities still face ordering and price impact, so this is allocation fairness rather than equal execution price. No privileged creator or agent exemption. At endBlock every buy and sell uses ordinary v4 routing with no proof, and the premium is permanently zero. The hook cannot later raise fees, extend the window, change the root, or move the cap. The vault has fixed distribution rules and sponsor payout cap; staker recipients vary only by voluntary deposits, with no hook/vault admin key or proxy. IMD's separate PoolFees.setRequester lets the current requester redirect future LP fees immediately, so the vault must be installed as requester before the first open swap and must have no function to forward that call; otherwise the no-admin rule is unmet. Test token ordering, opposite swap direction, flash routing, split bundles, exact-output rejection during opening, rejection of EIP-1271 signers, nonce reuse, replay across pools/chains and arithmetic overflow. IMD verified PoolFees source within LaunchFactory bundle.

5. Agent spec

What the agents do (Inference design, 2026-10-23): before launch, three independently owned IMD-seat agents inspect candidate seat ownership, ERC-8004 identity metadata, onchain funding/transfer clusters and prior accepted-work evidence. They use judgment to flag likely common control and publish a short evidence hash plus signed yes/no decisions for each candidate. Graph hints can be produced by ordinary code; deciding whether ambiguous profiles and work histories establish independent agents is the AI task. Agents submit their decision Merkle roots and evidence content hashes onchain to a nonupgradeable ReviewEscrow with three immutable reviewer EOA addresses and an Ethereum snapshot hash. On Sepolia, the escrow verifies their signatures, matching-root quorum, duplicate-owner/equivocation challenges and a fixed delay, then pays each valid reviewer from a prefunded balance. Ethereum ownership and WorkRegistry claims in the root are checked on a pinned Ethereum fork before the sponsor funds it; Sepolia cannot directly verify that other chain's state. A challenger can demonstrate two included leaves with the same owner and different keys, or conflicting signed roots; the escrow withholds the disputed bounty. Limitation: quorum and these mechanical checks cannot prove independence of humans, the truth of every Ethereum receipt, or the quality of subjective analysis, so the root remains a trusted-reviewer choice. This is the core economic experiment, not a trustless Sybil cure. IMD docs, ERC-8004 EIP.

Who signs what (Inference design, 2026-10-23): each pinned reviewer seat owner signs an EIP-712 (chainId, escrow, snapshotBlock, candidateRoot, evidenceHash, expiry) attestation and posts it onchain. The eligible EOA seat owner or registered EOA agentWallet signs each swap permit described in §4 and must be the transaction origin. The hook checks signature, origin, root proof, nonce and cap before any fill. The escrow checks reviewer signatures, two matching roots from the three immutable reviewer EOAs and onchain-provable duplicate/equivocation predicates, waits a fixed 24-hour challenge period, then pays ETH to each signer; it cannot check whether the AI's inference or an Ethereum-only receipt is true. The hook never pays a reviewer per trade, so reviewers have no incentive to approve extra volume. A missing or contested root prevents opening launch; it does not trigger an emergency admin override.

Oracle question (Inference, 2026-10-23): no oracle.request in the smallest MVP. If the escrow test succeeds, a separate optional check can ask one chain-local Boolean: “At Ethereum block N, did every ERC-8004 ID in the named candidate list have at least one non-revoked NewFeedback entry from the IMD WorkRegistry client, yes or no?” N, the candidate list, the WorkRegistry client and the exact feedback predicate must be fixed in the paid request; do not submit a vague question. IMD panels return one EIP-712 signature, checkable onchain, but the panel signer remains trusted and responses can take longer than opening blocks. The onchain escrow is the payment gate; an oracle response is corroboration, not the per-buy admission path. No oracle scheduling is used for the opening: official minima are 10 minutes for oracles and 30 minutes for jobs, far slower than the first blocks. IMD docs, capabilities.

Identity-farming measurement (Record, read 2026-10-23): the ERC-8004 EIP permits anyone to register identities and warns that reputation needs a trusted client context. One owner (0x28556846b891f349817f8731ad85cae5800acd86) registered five consecutive Ethereum IDs in blocks 26,254,241–26,254,293 on 2026-10-23: first, second, third, fourth, fifth. Each successful registration used 598,347 gas; effective gas prices were 56,213,970, 60,524,348, 60,342,648, 55,147,275, and 56,806,117 wei/gas. Total observed fee = 598,347 × (56,213,970 + 60,524,348 + 60,342,648 + 55,147,275 + 56,806,117) / 10^18 = 0.000172942841006226 ETH, or 0.0000345885682012452 ETH per ID on average. This excludes metadata hosting and any future reputation-building cost.

Identity conclusion (Inference, 2026-10-23): a bare agent ID is economically unsuitable as a one-per-person key. IMD seat-owner snapshot at 2026-10-23 19:45:14 UTC reported 2,000 seats, 788 owner addresses, and a largest owner holding 77; a seat is scarcer but still not one human. This is an owner-address count, not a beneficial-owner proof.

Reputation write check (Record, read 2026-10-23): IMD's feedback/batches lists a sent Ethereum batch with transaction 0x0d6a5212b354fc5e0195e5d7d3bc97cf8da670b49588f4b216331d9a14bd97ca in block 26,194,072, dated 2026-10-14 23:36:20 UTC. Its receipt succeeded and emitted 12 NewFeedback logs at the official Ethereum reputation registry, so the answer is yes: some IMD reputation writes landed onchain. The most recent API page also contains queued and failed batches; therefore an IMD work result is not automatically a landed reputation entry. The official deployment lists identity 0x8004A169FB4a3325136EB29fA0ceB6D2e539a432 and reputation 0x8004BAa17C55a88189AE136b182e5fdA19dE9b63 on Ethereum and Robinhood Chain; an official validation registry deployment is not found. ERC-8004 contracts, IMD docs.

Failure modes (Inference, 2026-10-23): colluding reviewers, bought or transferred seats, a single actor funding multiple agent wallets, false metadata, delayed or disputed reputation batches, signature replay, expired permits, and an oracle signer outage. Revoke a compromised root only before immutable hook deployment; after deployment there is no key and no rescue path. A failed root means do not launch. ZK nullifiers are excluded: they would conceal the buying wallet only after a credential issuer has already established independence, and would add proof cost without removing the trusted reviewer. ERC-8004 EIP, IMD docs.

6. Economics

Day-one payer (Inference, 2026-10-23): the launch sponsor funds IMD requests, deployment gas and an ETH ReviewEscrow before reviewer work. The buyer pays the normal pool fee and a small opening-only premium. The sponsor pays because the experiment seeks a defensible initial holder distribution; agents do not bear launch risk. The bounty is paid onchain from escrow after verification, independent of IMD's offchain compensation or EIP-712 oracle signature. Sepolia test ETH proves payout mechanics but is not economic compensation; a later public launch needs a real-ETH bounty accepted by the reviewers before they work. If no sponsor prefunds, this version should not launch.

Recorded unit costs (read 2026-10-23): IMD capabilities and docs price job.open, oracle.request and launch.open at 0.5 IMD each on Ethereum/Robinhood Intake. Two jobs plus launch = 3 × 0.5 = 1.5 IMD; a subsequent oracle = 4 × 0.5 = 2.0 IMD cumulative. The 10% swarm supply share is a token allocation, not 0.5 IMD cash. Gas, hook salt mining, reviewer bounties and legal/operational costs have no verified quote here. Record (read 2026-10-23): IMD's current pool LP fee is 1.25% and the requester/network division is 80/20 of that fee, or 1.25% × 0.8 = 1.00% and 1.25% × 0.2 = 0.25% of fee-bearing input before other v4 accounting. Verified IMD LaunchFactory, IMD docs.

Proposed split and break even (Inference, 2026-10-23): route the requester fee to an immutable vault: 80% of that share to opt-in token stakers and 20% to repay sponsor costs until a fixed cap, then 100% to stakers. Thus at current LP rates, the sponsor's maximum fee recovery is 1.00% × 0.20 = 0.20% of fee-bearing trading input and holder reward is at least 1.00% × 0.80 = 0.80%; the network keeps 0.25%. The opening 0.25% premium goes to holder rewards, never to the sponsor, so it cannot bias sponsor incentives to prolong the opening. Let C_ETH be actual sponsor gas + escrow bounties + the 1.5 IMD MVP requests converted at the paid execution price, and V_ETH be total fee-bearing input measured in ETH equivalents. Cost recovery needs 0.002 × V_ETH ≥ C_ETH, so V_ETH ≥ C_ETH / 0.002 = 500 × C_ETH; this omits price impact, changing fee bases and vault gas. Since no IMD/ETH price or launch volume is verified, a numeric break-even volume is not found. The fee design rewards holders while capping total opening fee near 1.5% rather than a 99% sniper tax, but volume retention remains an experiment. IMD docs, Pons v2.

7. Home fit

IMD (Record and Inference, read 2026-10-23): the platform already offers a custom v4-hook launch on Ethereum/Robinhood, a test chain, fixed-supply token rules, seat owners and paid work. Its swarm API reported 1,026 agents online, 1,055 enrolled seats, 69 jobs and 16 oracles completed in the preceding day at 2026-10-23 19:54 UTC; this is an API snapshot, not an independently audited labor count. The Ethereum LaunchFactory was Sourcify-verified 2026-10-08; Intake 0xa43e6f75ee006411f79ac1c84120606c2330de82 is documented on chains 1 and 4663. IMD must accept this specific hook permission mask and beforeSwap accounting, disclose/test the fee-recipient handoff, allow a Sepolia hook launch, and supply the sponsor with actual deployment parameters. A concrete no-admin path is a prefunded, immutable vault that submits the fixed launch.open request itself and is recorded as the initial requester; whether IMD accepts a contract payer as requester is not verified. The sponsor must commission/prefund review. IMD's LaunchFees owner can change the network fee recipient without a timelock, though the existing pool fee and fixed 80/20 LP split cannot be changed through it; strict interpretation of the no-admin rule therefore needs IMD to constrain that external control too. If the factory refuses the mask, fee accounting or recipient handoff, stop; there is no compliant workaround via token transfers or a venue lock. IMD capabilities, docs.

CLAUS (Record and decision, read 2026-10-23): the hook 0x37Bfb8AC7C960E558657871D41Ca70E07e7DbfFf is frozen: its EIP-1967 admin slot resolved to the contract that emitted ownership transfer to zero in block 26,197,788 at 2026-10-15 12:03:11 UTC, and that admin's owner() was still zero at block 26,257,408. The verified TradeExtensionHost 0xea0436465676eA7001535Bfe2E5cbd4b137b7f91 nevertheless lets the token owner immediately replace its module. The host passes each swap's router sender and hookData to a module and caps returned allocations at 149,000 ppm = 14.9%. The fee-state record at block 26,257,472 on 2026-10-23 reports a current total 23,000 ppm = 2.3%, zero direct liquidity allocation and a paused LP reserve; the published plan alone is not a promise of current LP rewards. CLAUS would need its owner to transfer module control to a timelock of at least 24 hours and approve a prospective new launch; this already-live token's first blocks cannot be reopened. If CLAUS says no, use IMD Sepolia; if IMD says no, stop this particular build instead of changing the hard requirements.

Hard-requirement audit (Inference, 2026-10-23): (1) agents make judgment-bearing onchain attestations and receive checked escrow payouts in the design, but reviewer quality is unproved; (2) the v4 callbacks and flags are specified in §4; (3) the ERC-20 is plain and the hook has no venue allowlist; (4) the proposed hook/vault have no admin key, but IMD's external network-fee setter needs a strict-scope decision and current CLAUS fails the 24-hour control rule; (5) bundle splitting is capped by identity, but identity farming is unproved; (6) LP fees and the vault reward liquidity/holders with a capped opening premium, but volume impact is unmeasured; (7) no ZK is required for MVP. These unresolved parts make a public launch a no under the seven-part rule, even if a Sepolia falsification test is worth buying.

8. Why now

Dated records and inference (read 2026-10-23): the last-60-day interval is 2026-08-24 through 2026-10-23. IMD's documented Intake v2 launch/workflow path went live 2026-10-10 to 2026-10-13; the verified LaunchFactory source was published 2026-10-08; an actual IMD ERC-8004 reputation batch landed 2026-10-14. These events make a test using real seat/work state possible now. IMD docs, Sourcify factory, feedback batch.

Dated records and inference (read 2026-10-23): CLAUS's host was verified 2026-10-15, giving a concrete example of hookData-aware fee logic but also a live example of the timelock gap. Host source. Uniswap v4-periphery commit 0c4e5bd3715ce2e72fd7609a1e2835f8300baf7b dated 2026-10-22 and v4-core commit bea7f6d1c11cd5dc4f0d5151ffa37a1c239feee6 dated 2026-10-23 fix the code pin for a reproducible hook test. The CCA repo also changed 2026-10-22, making the simpler alternative current. These are reasons to test promptly, not evidence that an unaudited launch can safely ship in two weeks.

9. MVP plan of 14 days or less

Plan (Inference, formed 2026-10-23; day counts are targets, not observations):

Days	Deliverable and gate
1–2	Freeze IMD factory, registry and v4 source pins; snapshot Ethereum seat owners and landed WorkRegistry receipts; choose one token cap and reviewer bounty from sponsor budget; publish attack model. The Sepolia test NFT is open mint and is not used as identity evidence.
3–5	Build AgentOpeningHook, ReviewEscrow and immutable fee vault; use IMD's standard fixed-supply launch token. Post three independent reviewer attestations to escrow; no admin, proxy, transfer controls or ZK.
6–8	Foundry unit and invariant tests against pinned v4 core/periphery: actual output cap under same-block bundles and many nonces, copied-permit front-run from a different EOA, price movement, token ordering, stale/forged EIP-712, rejection of ERC-1271/account-abstraction path during opening, reentrancy, root replay, exact-output opening, endBlock transition, fee monotonicity, permissionless premium sweep, staking reward accounting and escrow payout.
9–10	Fork tests against IMD factory and Uniswap router/Universal Router; trace hookData to PoolManager; try a representative aggregator. Measure gas and public quote availability. Reject any route that silently strips data.
11–12	Independent review of custom accounting and payout. Submit fixed yes/no oracle question only if reviewer receipts and snapshot make it objective.
13–14	If all kill gates pass and IMD accepts the hook, request Sepolia launch and run adversarial bundled buys; publish traces, gas, recipient behavior and fee accounting. No mainnet deployment is authorized by this plan.

Exact subsequent IMD orders (Inference, 2026-10-23): (a) job.open, software/build: “Implement and test immutable v4 AgentOpeningHook + ReviewEscrow with the §4 mask, canonical seat/agent key, real-output cap and fee-vault math on pinned v4 commits; deliver Foundry sources and traces.” (b) job.open, independent security review: “Reproduce split-bundle, router-sender, custom-accounting and payout tests and sign a yes/no finding for each stated invariant.” (c) launch.open, target Sepolia univ4_hook, pointing to the audited hook and IMD's standard fixed-supply token, only if the factory accepts it. Order (a) and (b) through Ethereum or Robinhood Intake and (c) through a supported Intake/API path targeting Sepolia; do not send launch.open to Sepolia Intake, which sells only oracle questions. Each order costs 0.5 IMD per capabilities; no order is placed by this research task. The optional §5 oracle is a separate later decision, not an MVP order. IMD docs.

10. Kill criteria

Measurable gates (Inference, set 2026-10-23):

1. Identity economics: in a documented exercise, if a single funding/owner cluster can obtain two or more opening allocations for a cost below the value of one allocation, kill the mainnet proposal. Record gas and actual cap value at the test snapshot; a bare ERC-8004 registration already fails the identity standard. ERC-8004 EIP.
2. Accounting and bundle safety: any Foundry fuzz/fork counterexample with filled[key] > cap, opening premium above 2,500 ppm, a post-opening premium above zero, a sell blocked by the gate, or a fee-vault shortfall kills this hook implementation. v4 core.
3. Route availability: if fewer than two independently controlled transaction paths pass valid hookData and produce a normal user quote on Sepolia, or if a required integrator does not allowlist the hook, kill a public launch. Direct v4 router plus a custom transaction builder count as only one independent path. Uniswap allowlist.
4. Agent payment and independence: if three distinct seat-owner addresses cannot submit eligible decisions, the escrow fails to pay all valid reviewers after its fixed delay, an independent Ethereum-fork check finds any false receipt claimed in a paid root, or the reviewers will not accept the sponsor's posted real-ETH bounty terms for production, kill the “agent” claim. Beneficial independence remains an unresolved audit item even if addresses differ.
5. Home admission: if IMD factory rejects the hook address, delta accounting, fee recipient or Sepolia launch, kill this deployment rather than adding a transfer tax or privileged router. IMD factory.

11. Moonshot

Boldest compliant version (Inference, 2026-10-23): make every launch on IMD use a private, portable one-per-independent-agent credential whose zero-knowledge nullifier is bound to a single pool and opening window. The proof conceals the registry ID and its link to the buying wallet while the hook checks a nullifier and cap; the router or settlement transaction may still expose the wallet itself. A separate threshold of economically bonded AI reviewers signs credential eligibility and is paid by an onchain escrow for evidence-backed decisions. The token remains freely transferable, the pool becomes ordinary after the opening, and the hook/vault have no admin. This ZK layer removes the trusted linking service that otherwise learns both identity and buying wallet; it does not remove trust in credential issuance or reviewer judgment. ERC-8004 EIP, v4 Hooks library.

Smallest falsifiable experiment in 14 days (Inference, 2026-10-23): on a local Foundry fork, issue three real registry-backed credentials to distinct test agents, make two wallets per agent produce proofs, then show one accepted opening buy per agent, a second-wallet rejection with the same nullifier, no public link from swap calldata to registry ID, and a reviewer payout onchain after a valid decision. Measure prover time, proof calldata/gas and a deliberately duplicated credential. The sponsor pays prover/build gas and prefunds reviewer ETH bounties on day one; proposed pool fees can reimburse only after real volume. No chain launch is required for this falsification.

Why it might be a bad idea (Inference, 2026-10-23): ZK can hide only the wallet link, while an attacker can still farm ERC-8004 IDs or control multiple seats and receive multiple valid credentials. The trusted reviewers become a bottleneck and may collude; proof generation adds latency exactly when a launch is most time sensitive; opaque eligibility and missing aggregator support may exclude ordinary buyers. CCA's equal-price auction or a simple time-decaying fee may supply enough practical fairness without identity issuance, reviewer disputes or a bespoke cryptographic circuit. ERC-8004 EIP, CCA README, Uniswap hook routing.

12. Open questions and everything not verified

Unknowns (read 2026-10-23): IMD Sepolia admission of the exact 0x20c8 mask and beforeSwap custom-accounting delta; fee-recipient timing, contract-payer acceptance and whether the immutable vault can be the initial requester; whether IMD's external network-fee setter counts under the no-admin rule; whether IMD jobs can recruit specific reviewer seats and deliver individual signed attestations; whether reviewers will accept real-ETH bounty terms; exact ERC-8004/seat-owner adapter ABI on Robinhood Chain; historic beneficial ownership of seats; official validation registry deployment (not found); support for hookData in 0x, 1inch and other aggregators (not found); a safe recipient-verifying router path for smart accounts and relayers (not found); actual router output recipient from arbitrary routing; whether Uniswap will allowlist this hook; a quote for gas, bounties, IMD/ETH conversion and trading volume; the qualitative error rate of AI Sybil review; independently dated Pons v2 release activity (not found). A Sepolia proof tied to an Ethereum snapshot cannot show that a transfer after the snapshot is reflected on Sepolia. The existing CLAUS owner agreeing to a timelock or a new launch is also unverified. These are launch gates, not values to invent.

Interpretation (Inference, 2026-10-23): no address in the proposal is a deployment target until its exact bytecode, ownership and chain are checked at a pinned block. Official ERC-8004 registry addresses above are verified against the official deployments and observed code; the proposed hook, escrow, vault and token addresses are not found because nothing has been deployed. The local RPC reads used for gas and receipt checks are cited through the public transaction links and official contracts, never as a publishable endpoint.

13. Sources with read dates

All sources below were read on 2026-10-23 UTC. Onchain values are pinned above to blocks/transactions; API values are time-stamped above. GitHub repository state is pinned below where the full hash was retained. A page owned by a product is treated as its claim unless corroborated by contract source or a transaction.

• IMD: docs, capabilities API, swarm API, seat-owner API, older sent feedback batch, recent batch queue, Ethereum LaunchFactory verified source (Sourcify verified 2026-10-08 17:44 UTC), Robinhood LaunchFactory verified source (Sourcify verified 2026-10-08 08:55 UTC), landed reputation transaction (block 26,194,072).
• Identity: ERC-8004 EIP, official contracts and deployments at commit b9e466c250744a7e06b13dff9d3c2844ed64f825 dated 2026-08-15; same-owner registration receipts one, two, three, four, five (blocks 26,254,241–26,254,293, 2026-10-23).
• Uniswap v4: core Hooks library, IHooks, PoolManager, router, router interface; core pinned at commit bea7f6d1c11cd5dc4f0d5151ffa37a1c239feee6, 2026-10-23; periphery pinned at commit 0c4e5bd3715ce2e72fd7609a1e2835f8300baf7b, 2026-10-22. Hook allowlist and routing guidance.
• CLAUS: hook verified source, proxy-admin renunciation transaction (block 26,197,788, 2026-10-15), TradeExtensionHost verified source (verified 2026-10-15), fee-state JSON (observed 2026-10-23 19:39:09 UTC, block 26,257,472).
• Competitors: Pons v2 official docs and factory event (block 96,196,144); Flaunch API docs and SDK commit f9bc6ac58d1d3c16f7899919a9bec6a2d9e81ecb dated 2026-10-21; Doppler explainer and SDK commit 3eaa03e37057b304e81e0fc61a12612c3f5ecf91 dated 2026-10-20; Uniswap CCA README, technical docs, commit ce78b4d94cdf94fd41f20afb6f6e495b8b83a234 dated 2026-10-22, Uniswap auction announcement dated 2026-01-28; Zama auction event dated 2026-01-26 and auction docs.