1. Question restated

Is Idea D, Open Lane, worth building and launching within 14 days on IMD or CLAUS, and, if so, which of three distinct ideas is the smallest credible proof?

2. Verdict in five lines

Gap: Partly served; yes to a gated IMD pilot. V4 fee defenses and fee-funded rewards exist, but an LP-approved, agent-authored range plan with payment only after onchain execution is not established by the sources below (Inference, 2026-10-23). Speed to MVP: 14 days, conditional on a factory fork proving that its requester fee can be collected by the hook and split in both pool currencies (Inference, 2026-10-23). MVP: IMD LP Desk on Robinhood Chain: one fixed-supply token, one v4 hook, a pro-rata opening batch, one managed LP position, one signed agent plan, and a fee-funded onchain agent payment (Inference, 2026-10-23). Biggest reason it could fail: no LP other than the sponsor may value qualitative range advice enough to accept the plan, while the factory's existing 1.25% trading fee already competes with cheaper pools (Inference, 2026-10-23; IMD launch rules). Total IMD cost: 1.5 IMD = three orders × 0.5 IMD, after this research order; gas, liquidity, and agent operating cost are additional and unpriced (Record for unit price, Inference for proposed order count, 2026-10-23; live capabilities).

3. Competitor table

Ranked proposals. 1 — LP Desk (winner): an AI agent reads unstructured project releases, governance discussions, and competing venue conditions, writes a sourced LP range and expiry, signs it, and executes the accepted plan through a managed vault. The v4 hook recognizes only that signed plan for the managed position; after execution and fee receipt it pays the agent onchain and sends the remaining requester fees to the LP vault. A timed pro-rata opening batch removes transaction-order advantage. No token-level controls, owner-free hook, 24-hour timelock for the order payer, and no ZK. Trading fees never order recurring IMD jobs, the opening rule is deterministic rather than agent-priced, and the pool is on IMD rather than CLAUS; those are the boundaries from Fee to Jobs, Agent First Opening, and Claus Crew. This is the only proposal here whose work quality is purchased directly by the LP rather than inferred from an oracle (Inference, 2026-10-23).

2 — Catalyst Reserve (short version): agents read unstructured official disclosures and contract changes, then submit a signed yes/no claim onchain before a panel is ordered. A sample pinned question is: “Did the issuer publish an official notice at the specified URL, before the specified closing block, announcing a pause of the named contract?” The v4 hook uses afterInitialize, beforeSwap, and afterSwap to run the same opening batch, record fee-bearing trades, and direct a fixed portion of the factory's existing requester share to LPs or a reserve according to a fresh IMD EIP-712 panel attestation. A matching claimant receives a fixed portion onchain only after questionHash, chain, expiry, signer, and quorum checks and actual requester-fee receipt; a fixed share still goes to LPs. The token is ordinary fixed supply, the hook has no owner setters, payer rights are behind a 24-hour timelock, bundles in the opening get the same pro-rata price, and no ZK is used because it would not remove the panel signer. A sponsor would pay for the first panel and opening liquidity; the smallest test is one resolved claim, one independent LP, and one nonzero fee receipt within 14 days. Its proposed orders are payer deployment, agent job, hook launch and one oracle request: 4 × 0.5 = 2.0 IMD; a missing panel answer or a payment without a valid signature kills it. It ranks lower because the panel signer remains a trusted party and a subjective question could disagree (Inference, 2026-10-23; oracle format).

3 — Fee Stream Underwriter (short version): agents read offchain demand and project evidence, submit signed bids onchain and put their own capital into a uniform-price auction for a fixed future requester-fee stream. They receive only the actually collected stream onchain; the escrow pays auction proceeds to an LP reward vault. The v4 hook uses afterInitialize, beforeSwap, and afterSwap to enforce the opening batch and immutable stream period. A sponsor would pay the first orders and LP seed, while bidders pay the fee-stream price; 3 × 0.5 = 1.5 IMD covers payer deployment, an agent job and a hook launch. A 14-day test needs a genuine competing agent bid, an independent LP and one claim from collected fees; an empty auction or any unearned payout kills it. The token has no transfer controls or later mint, the hook has no owner key, payer routing uses a 24-hour timelock, bundles cannot change the batch opening price, and ZK adds no trust reduction. It ranks third because Tenet already proposes much of this on Robinhood Chain and is still pre-launch by its own account (Claim, read 2026-10-23; order counts are Inference).

Name	Link	Chain	Status	Last activity, pinned commit	Overlap	What it misses relative to LP Desk
BRO-AMM	Repository	Ethereum, Arbitrum data	unreleased	2026-10-15, e9fd5dad (Record)	LP loss recapture	No LP-selected semantic plan or onchain agent reward.
Hindsight	Repository	Unichain Sepolia	testnet	2026-09-03, 63efd09c (Record)	Ex-post LP fee settlement	Scores markout by code; no agent reading qualitative evidence.
Uniswap VolatilePairHook	Repository	Ethereum	live	2026-10-23 repository commit 2264ac8b (Record)	LP fee adaptation	Algorithmic fees; no paid LP plan author.
Pines	Repository	Robinhood Chain	live	2026-10-05, cddf9768 (Record)	Fee-to-staker stream and launch snipe tax	Its 90% opening buy tax for 30 seconds (Claim, read 2026-10-23) differs from equal-price batching; no paid LP plan author.
Even	Repository	Monad	live	2026-10-14, 5c316b59 (Record)	Uniform-price fair launch	Uses a v3 adapter; no continuing LP strategy market.
Tab	Repository	Monad	live	2026-10-18, f4cdd36d (Record)	Onchain agent payment accounting	Postpaid service billing, no v4 hook or LP yield.
Obelisk	Repository	Robinhood Chain	live	2026-09-25, 07d27bb8 (Record)	AI agent onchain execution	ZK spending policy, no LP plan or hook fee distribution.
Veil	Repository	Robinhood Chain	live	2026-10-15, e8626b63 (Record)	v4 hook and ZK	Privacy and gated swaps; no LP-accepted agent plan.
Tenet Agents	Repository	Robinhood Chain	unreleased	2026-10-05, ce941f5b (Record)	Agent pricing of future creator fees	Stream underwriting, not LP range execution.

The status labels for projects other than Uniswap are their repository claims; a README or deployed address alone is not independent evidence of sustained user volume. Obelisk calls itself beta, Even says its mainnet rounds were scripted, and Veil's exact mainnet scope was not independently verified (Claims/uncertainty, read 2026-10-23). GitHub's commit metadata was read on 2026-10-23, and every hash above pins the reviewed tree. The cited Uniswap source was also pinned at commit bea7f6d1 on 2026-10-23 (Record).

4. Hook spec

Home and opening. Use launch.open with onchain: "univ4_hook", chainId: 4663, pairWith: "eth", ordinary IMD factory token, and a hook whose constructor creates its opening auction and single-position vault. The factory currently admits hook launches on Ethereum and Robinhood Chain, makes a standard 1,000,000,000-token, 18-decimal plain-transfer supply, gives the swarm 10% of it, mints the supply once, and takes a 1.25% pool trading fee: 1% to payer and 0.25% to network (Record, read 2026-10-23; docs, capabilities). The project requests no post-launch mint, transfer fee, transfer restriction, forced venue, or extra hook swap charge (Inference/design, 2026-10-23).

Callbacks and flags. afterInitialize (0x1000) binds the one factory pool key, opening deadline, and auction; beforeSwap (0x0080) allows only the auction contract to make the aggregate opening buy until settlement/refund and accepts all ordinary routers only afterward; afterAddLiquidity (0x0400) records a vault position only after both signatures, exact ticks, salt and liquidity delta verify; afterRemoveLiquidity (0x0100) ends that position's eligibility without ever blocking an exit. The permission mask is 0x1000 + 0x0080 + 0x0400 + 0x0100 = 0x1580 (Inference/design, 2026-10-23). All other permission flags, including every returnsDelta flag, are false. These flag definitions and callback arguments are in v4-core Hooks.sol and IHooks.sol (Record, read 2026-10-23).

Hook data and identity. Ordinary swap hookData is ignored; no trader is metered, so a router sender is never mistaken for a trader. During opening, beforeSwap requires sender == auction and the auction's fixed auctionId; a depositor's address is resolved by the auction escrow, not by the v4 callback. Managed liquidity data is abi.encode(uint8 version, bytes32 planHash, uint256 nonce, bytes agentSignature, bytes lpSignature); only sender == vault can register it. The signed EIP-712 plan binds chain ID, PoolKey hash, vault, agent payout wallet, LP beneficiary, lower/upper ticks, maximum deposit, nonce, expiry, and evidence hash. Anyone can provide ordinary LP liquidity without a plan; the hook returns zero delta. A v4 hook receives the caller of PoolManager (usually a router), as shown by the core call path (Record, read 2026-10-23).

Opening fairness. Proposal constant: a 24-hour escrow window after initialization (Inference/design, 2026-10-23). Buyers deposit the quote asset during that window; after it, anyone calls settle, which makes one aggregate pool buy with an immutable sqrtPriceLimitX96 published before deposits. Settlement requires the whole aggregate input to clear within that limit; a partial fill reverts. The hook remains in auction-only mode until settlement succeeds or a fixed refund timeout completes, so a bundle cannot jump ahead at the deadline. Each depositor receives floor(totalTokensBought × theirDeposit / totalDeposits) (Inference/formula, 2026-10-23); unclaimed rounding dust goes to the LP vault, with no last-claimer bonus. The same average execution price and proportional rounding apply regardless of transaction order. Bundling deposits changes capital weight only; splitting wallets gains nothing. The price limit bounds a settlement sandwich; if it cannot be met, the timed fallback refunds all deposits and opens ordinary swaps, and that failure is a launch kill event. Token transfers and independent pools remain possible throughout, so this is a rule of this pool's opening, not a token venue lock. The batch removes official-pool queue advantage, not all external-market or LP manipulation; the fork tests must check both. As Even shows, equal-price opening is an existing concept; this version makes it a small launch component (Inference, 2026-10-23; v4 SwapParams).

Fee math and distribution. The payer is a timelocked order contract; after launch it calls the documented setRequester(launchNumber, hook) so the hook receives the existing 1% requester stream in both pool currencies. The hook has a native-currency receiver and token-safe claim path; a reentrancy guard and per-currency balance deltas separate actual fee receipts from principal. A permissionless claim/distribution function books those received balances, sends 80% to the funded LP vault and at most 20% to the plan agent, and sends the agent's unused portion to the LP vault. The 80/20 is a proposed immutable allocation, not observed demand (Inference/design, 2026-10-23). Worked example, separately for either currency: on 100 units of fee-bearing gross trade, the factory takes 1.25 units = 1.00 requester + 0.25 network; of 1.00, funded vault receives 0.80 and a qualifying agent can receive 0.20. Thus 0.80 + 0.20 + 0.25 = 1.25 and the proposed hook adds zero swap fee (Record for factory rates, Inference for split and arithmetic, 2026-10-23; factory policy). A real buy/sell can deliver different currencies and rounding; the worked example does not imply automatic conversion.

Invariants and admin. Agent payment cannot exceed 20% of collected, not merely quoted, requester fees; one plan nonce pays once per distribution epoch; no agent payment before valid LP consent, actual position creation, and a nonzero post-activation fee receipt; no payout may consume principal; a failed plan leaves 100% of that receipt for the LP vault. afterRemoveLiquidity cannot revert because of a stale/missing plan. Hook/auction/vault fee rules are immutable and have no admin key. The order payer's setRequester, funds withdrawal, and any factory-facing powers belong only to OpenLaneTimelock, with an immutable delay floor of at least 24 hours and no shortening path (Inference/design, 2026-10-23). The factory fee payout ABI and recipient compatibility must be verified in a fork before an order to deploy; the public docs state the right and both-currency receipts but do not publish verified factory source (IMD payer documentation, read 2026-10-23).

5. Agent spec

An open agent reads qualitative information a deterministic range algorithm does not understand reliably: project release text, governance actions, liquidity migration notices, and contradictions between venue announcements. It produces a source-hashed memo and bounded tick range, signs the EIP-712 plan, and submits the plan hash in an onchain transaction before offering the executable plan to an LP. This is onchain work because the agent commits to and authorizes a specific v4 liquidity operation; a plain price-band formula can calculate ticks but cannot reliably interpret conflicting prose or source credibility (Inference, 2026-10-23). A second, independent agent may compete, but only the LP-selected plan is eligible for payment.

The LP funds the vault and signs the exact plan hash and bounded vault authorization. The agent sends executePlan onchain with that authorization; the vault calls PoolManager. The hook validates the prior onchain agent commitment, both signatures, deadline, nonce, PoolKey, ticks, salt, liquidity delta, and vault sender in afterAddLiquidity. A successful callback plus the PoolManager position change verifies execution; the fee collector's actual balance change verifies that revenue exists before anyone is paid. The agent withdraws its credited currency onchain. The LP can remove liquidity at any time; exit ends future agent eligibility. Neither an agent's self-assertion nor a project website alone releases funds (Inference/design, 2026-10-23; v4 callback contract).

Pinned oracle question: none for the winner. IMD's oracle does return one EIP-712 signature covering a panel result, so it is useful for the alternative Catalyst Reserve, but that signature is still a signer trust assumption; no ZK proof improves this MVP's LP-controlled acceptance (Record, read 2026-10-23; IMD attestation specification). The official ERC-8004 repository lists Identity and Reputation Registry addresses on both chains but no official Validation Registry deployment; its latest commit is 2026-08-15 (Record, read 2026-10-23). Agent IDs can be logged as optional metadata, never as a payout authority.

Failure modes: biased or copied memo, collusion with an LP who signs a useless plan, stale news, reorgs, replayed signatures, false fee accounting, fee collection failure, and opening escrow failure. The controls are evidence hashes and public comparison, LP consent with cancellable nonce, expiry/domain binding, finalized balance accounting, and a full refund path for the opening. The design cannot prove an AI plan was optimal; the pilot tests willingness to buy and verifies payment integrity, not alpha (Inference, 2026-10-23).

6. Economics

On day one the launch sponsor funds the opening pool and voluntarily gives up its 1% requester fee stream to attract LP depth; the first LP must sign and fund a position. The sponsor pays the initial IMD orders. The LP accepts a plan because any reward and improved range income must outweigh management risk; this remains unproven demand. An agent earns from fees only when a chosen plan is executed and the pool actually collects a fee, so no presale buyer is promised yield (Inference, 2026-10-23). The live IMD swarm snapshot at 2026-10-23 19:44:34 UTC showed 1,026 agents online, 69 jobs and 16 oracles completed in the prior day, and 807 live launches (Record, dynamic snapshot). Capacity is present; qualified LP buyers are not established by those counts.

Unit costs: each job.open or launch.open order is 0.5 IMD, and a scheduled run costs 0.5 IMD per run; jobs cannot be scheduled faster than every 30 minutes and oracles no faster than every 10 minutes (Record, read 2026-10-23; capabilities, schedule docs). The MVP orders one evm_contracts launch for timelock/payer, one job.open for an agent plan generator and pilot memo, and one univ4_hook launch with independent review: 3 × 0.5 = 1.5 IMD. No oracle or schedule is necessary for the winner. Gas, inference, liquidity capital and spread are unknown, so there is no defensible fiat total (Inference, 2026-10-23).

Break-even is a measurable inequality, not a fabricated forecast. Let V be fee-bearing volume in a single quote currency, P the quote-currency price of one IMD at payment, G_s sponsor gas, G_a agent gas, and C agent compute cost. Under the proposed split, maximum agent receipts are 0.002 × V because 1% × 20% = 0.2%; the MVP breaks even for its sponsor only if the value of LP retention/fees exceeds 0.01 × V + 1.5 × P + G_s, and for an agent paying its own costs only if 0.002 × V ≥ C + G_a (Inference/formulas, 2026-10-23). A later paid IMD job per plan adds another 0.5 × P to the sponsor's side; the MVP orders no such recurring job. The sponsor's 1% is a real opportunity cost; the hook does not impose another trader fee. Observe actual token prices and gas at execution before calling this profitable.

7. Home fit

IMD is the home. Current capabilities admit univ4_hook and evm_contracts on chain 4663 and chain 1, and docs permit fixed-supply standard-token hook launches and setRequester after deployment (Record, read 2026-10-23). The Intake address 0xa43e6f75ee006411f79ac1c84120606c2330de82 is documented on both chains; read-only eth_getCode returned nonempty code at Ethereum block 26,257,514, hash 0xf06d44a68db62d994febf9363ec2200043502bdb0a55263a0ea5b408aef39501, and Robinhood Chain block 96,193,417, hash 0x7bde1d362117d4d9730582b4b74e4e17a7fbbdfb0f1e219f9fe1793f46b3184c, on 2026-10-23 (Record; Ethereum Intake, Robinhood Intake). Code presence verifies the addresses are contracts, not that unpublished source implements every documented behavior. The same reads found code at the official ERC-8004 Identity and Reputation Registry addresses on both chains (Record; official deployment list, read 2026-10-23). Validation Registry address: not found in that official list.

IMD must accept the free request check for both launch bodies, support the hook constructor's auction/vault creation and 0x1580 permissions, and allow the requester stream to reach the hook. The payer must fund the orders, publish the real timelock owner at quote time, and route fees via a delayed transaction. If the factory cannot do this, the idea remains a local/fork prototype and does not launch under this verdict. The 10% swarm allocation and policy fee are accepted as fixed constraints, not negotiable assumptions (Inference, 2026-10-23; launch policy).

An IMD review of a separate Robinhood fee vault reports that it claimed requester fees and passed fork tests, while also finding a refused-launch state trap and fee-claim failure path (Claim scoped to that vault, read 2026-10-23). This supports testing the route, not assuming it works for a custom hook; the MVP must handle an Intake refusal and must not count a failed collection as a paid epoch (Inference, 2026-10-23).

CLAUS is not the default home. Its live fee state at block 26,257,500, hash 0xfd7051c749da39f5e8ba98a1b263c03456da2074fd6a67c5a3e31180785911ba, observed 2026-10-23 19:44:42 UTC, reports the main hook 0x37bfb8ac7c960e558657871d41ca70e07e7dbfff, TradeExtensionHost 0xea0436465676ea7001535bfe2e5cbd4b137b7f91, an active module, 23,000 parts per million (2.3%) current total fee, and a 149,000 parts per million plan maximum (14.9%; Record; percentage conversion pips/1,000,000). A read-only code check at Ethereum block 26,257,514 found code at hook and host, and activeModule() matched the feed (Record). The two CLAUS proxy admins' owner() changed from a nonzero address at block 26,197,787 (2026-10-15 12:02:59 UTC, hash 0x555e7b81bbc7ecc8775d5aada1300d79c90189d8677f735538b61d93e3607322) to zero at block 26,197,788 (2026-10-15 12:03:11 UTC, hash 0xa0c545a523c6a97f4459fd38e42dad74a3682b2dcf4802d9236248a96d33ce4d), still zero at 26,257,514: this verifies the October 15 freeze of those upgrade-admin paths (Record; block, token proxy admin, hook proxy admin). CLAUS's own state description says the core upgrade path is unavailable and independent pools remain permitted (Claim, read 2026-10-23). An independent IMD review of the candidate host describes the module receiving each swap's PoolManager sender and hookData and returning a bounded fee-allocation plan; it reports that the token owner can change a module immediately and that sender is usually a router (Claim tied to candidate code, read 2026-10-23). I could not verify a live 24-hour timelock on the current owner or that the host can observe LP add/remove callbacks. Thus this MVP cannot meet its admin and LP-verification rules on CLAUS without a new approved mechanism; if CLAUS declines it, proceed only with IMD (Inference, 2026-10-23).

8. Why now

The last 60 days before 2026-10-23 start on 2026-08-24 (Inference/date subtraction). The relevant change is the convergence of live hook infrastructure with narrow competing implementations: Hindsight's 2026-09-03 testnet tree demonstrates ex-post accounting; Obelisk's 2026-09-25 tree shows signed autonomous-agent execution with policy proofs; Pines' 2026-10-05 tree shows fee-to-staker plumbing on Robinhood Chain; Even's 2026-10-14 tree shows live-mainnet fair-opening claims; and Uniswap's 2026-10-14 VolatilePair announcement confirms production LP fee competition (Claims or Records as indicated in §3, links read 2026-10-23). None establishes this LP-consented AI payout, so the gap is partly served, not empty (Inference, 2026-10-23).

The practical timing is IMD's published 2026-10-10 Intake v2 and 2026-10-13 sale of jobs/launches plus the live 2026-10-23 capability response. IMD's control-plane version read 2026-10-23 was commit 962b3018e66f6ed5471ecf947a26101596c3fa61 (Record). Those dates make a 14-day build orderable; they do not prove users will buy it.

9. MVP plan of 14 days or less

Days 1–2: specify PoolKey, EIP-712 plan, immutable opening deadline, timelock roles, 80/20 accounting in both currencies, and the exact payer-to-hook fee route. Submit free POST /requests/check bodies before paying. Order launch.open@launch-1, onchain: "evm_contracts", chainId: 4663, contracts: ["OpenLaneTimelock", "OrderPayer"], shape: "chain", steps: [{skill: "build-contract-project"}, {skill: "adversarial-review"}], objective: deploy a 24-hour-minimum timelock and a payer controlled by it, with requestWithFailure, delayed setOrder, and delayed setRequester forwarding. The payer's constructor uses only static arguments; the future launch body is stored by the timelock after deployment, since this factory does not accept a dynamic bytes constructor argument. The real owner/proposer addresses are supplied by the sponsor at quote time; no address is guessed here. Cost 0.5 IMD (Inference/order plan; IMD contract launch docs, read 2026-10-23).

Days 3–6: order job.open@job-1, skill: "implement-and-test", github: true, objective: create an agent that reads specified public project/governance/venue sources, emits a cited bounded LP range and EIP-712 plan for the eventual agent wallet to sign, and produces a reproducible pilot memo. It must never hold LP funds or a worker's private key. Cost 0.5 IMD. Separately build the hook, its constructor-created auction and vault, and a testable fee adapter specification (Inference/order plan, 2026-10-23). The deployed runtime agent signs and submits its own plan onchain, then is paid by the hook's collected fee stream when its LP-approved plan executes; the IMD job charge is a network build expense, not the runtime agent's payout.

Days 7–10: Foundry unit/fuzz/invariant tests: hook address permission bits equal 0x1580 and the getHookPermissions() struct matches; foreign PoolKey rejected; factory seed allowed; auction-only first-window swaps and unrestricted post-settlement swaps; two deposits in opposite ordering get the same pro-rata fill; bundles/split accounts get no ordering advantage; settlement sandwich or JIT liquidity cannot bypass the fixed price limit; auction failure refunds; expired/wrong-chain/replayed agent or LP signatures fail; arbitrary router hookData cannot forge an LP; ordinary LP add/remove always works; for each currency agent + vault ≤ collectedRequester while the separate network share stays at the factory; no principal enters payouts; zero receipt means zero agent payout; and removing an LP position cannot be blocked. Test Intake refusal and a failed fee-collection attempt without losing the right to retry. Fork tests must use the real Robinhood factory/PoolManager and exercise setRequester, a buy, a sell, collection in both currencies, and a full LP exit. Secure a conditional commitment from at least one LP unrelated to the sponsor before the paid hook launch order. Foundry tests are necessary because v4 deltas and factory payouts are the main integration risks (v4 core, IMD launch policy, read 2026-10-23).

Days 10–14: schedule the payer timelock's batched setOrder plus order() on day 10, only if those tests pass and sponsor/LP funding is real. After its 24-hour wait, execute it by day 11 to order launch.open@launch-1, onchain: "univ4_hook", chainId: 4663, pairWith: "eth", shape: "chain", steps: [{skill: "build-contract-project"}, {skill: "adversarial-review"}]; set the launch owner field to the deployed timelock while generating no hook owner mutators, and specify standard fixed token economics, no added swap fee, auction/vault constructor and the verified callback flags. Cost 0.5 IMD. Settle the 24-hour opening on day 12; schedule setRequester after deployment and execute after its delay. On days 13–14, an LP accepts one real agent plan and funds one vault position, the agent executes it, and the pilot records at least one nonzero fee distribution. If the order or deployment cannot finish by day 11, the 14-day verdict fails rather than moving the proof date. Publish the fee distribution and opening receipts. If any prelaunch gate fails, stop before the paid deployment order. No contract is deployed or order placed by this research report (Inference/order plan, 2026-10-23; IMD hook example).

10. Kill criteria

• Integration failure: a fork test cannot route and collect the factory's documented 1% requester share in both currencies to the hook, or one add/remove path traps principal. Binary test result; stop before univ4_hook order (Inference/decision threshold, 2026-10-23).
• Fee integrity failure: any Foundry fuzz or fork trace shows total payout above collected requester fees, agent payment without a valid executed LP plan, an extra swap fee, or a token transfer/mint restriction. One counterexample kills launch (Inference/decision threshold, 2026-10-23).
• Opening fairness failure: either reversing two opening deposits changes their tokens-per-quote ratio beyond integer rounding, or a bundle of split deposits improves that ratio. One counterexample kills launch; an auction settlement refund on the real pilot also kills the opening design (Inference/decision threshold, 2026-10-23).
• Demand/volume failure: by day 14 there is no funded, LP-signed plan from a party other than the launch sponsor, no nonzero swap by a trader independent of sponsor and agent, or no agent has received a nonzero onchain fee payout. The pilot records independent wallet control offchain; the hook itself does not infer a trader from router sender. Any of these results kills launch-scale expansion even if the contracts work; these are proposed pilot thresholds, not market statistics (Inference/decision threshold, 2026-10-23).

11. Moonshot

The boldest extension is a permissionless market of competing AI LP managers: any agent commits a sourced range plan and an objective prediction of fee income versus a passive control, LPs sign only the plan they prefer, and the same hook pays agents from the realized surplus of their managed positions while preserving an immutable floor for LPs. It reuses the winner's exact callbacks, fixed-supply token, fair batch opening, no transfer or venue controls, no extra swap fee, no hook owner, and 24-hour-timelocked payer; ZK is omitted because it removes no trusted party here. The agent can read prose, governance, and venue events; the payout formula and baseline must be fully onchain. This remains a proposal, not a claim of achievable alpha (Inference, 2026-10-23). BRO-AMM's October 15 research tree and Uniswap's October 14 live volatile hook are competing baselines to beat, not endorsements.

Smallest falsifiable experiment under 14 days: on one pinned Robinhood Chain fork and one real pilot pool, run two independently signed agent plans against one passive full-range control for the same observation window, settle identical-fee accounting, and publish each position's principal-adjusted return and every source hash. Require at least one independent LP to choose a plan, and pay only from observed surplus; otherwise the moonshot fails. The sponsor pays the first IMD job.open at 0.5 IMD and funds any gas/LP capital; agents receive only onchain surplus shares (Inference/design and cost derivation, 2026-10-23; IMD price).

It might be a bad idea because concentration range performance is dominated by price path, inventory risk and selection bias; a short trial can reward luck or gaming of the control. An agent can also make a persuasive but false narrative, and LP consent does not make the advice safe. A complex surplus formula increases gas and attack surface, while existing fee hooks already improve LP economics with plain algorithms. If a robust matched control cannot be specified and independently reproduced, do not sell performance claims (Inference, 2026-10-23; Hindsight's own accounting limits).

12. Open questions and everything not verified

• The factory's deployed source is not explorer-verified according to IMD's own docs. I verified Intake code presence on both chains and hook/host code presence on Ethereum, but not source-to-bytecode equality or the factory's exact requester collection ABI. The latter is a launch gate, not an assumed implementation (Record/uncertainty, 2026-10-23).
• IMD factory and PoolManager governance powers, including any ability to alter the policy fee after launch, were not source-verified. The proposed no-admin hook / 24-hour-timelocked payer describes the new project's controls, not a claim that the wider platform has no administrators (uncertainty, 2026-10-23).
• Whether IMD's univ4_hook launch accepts a hook constructor that creates an auction and vault, and the exact factory callback order and gas budget: not verified. Free POST /requests/check, local deployment simulation, and fork tests must answer it before payment (uncertainty, 2026-10-23).
• The deployed CLAUS owner contract's current minimum delay and its authority over setModule: not verified. The candidate audit says immediate module changes were possible at its pin; the current owner may differ. No CLAUS implementation is recommended without a current code/role/timelock proof (uncertainty, 2026-10-23).
• The official ERC-8004 repository lists Identity and Reputation Registry deployments on Ethereum and Robinhood Chain, but no official Validation Registry address: not found. Identity is optional in this design (Record, official repository, read 2026-10-23).
• Competitor adoption, actual trading volume, costs of agent inference, gas, IMD exchange price, available LP capital, and a willing independent LP: not verified. The repository statuses in §3 are claims except Uniswap's live announcement. The break-even formula stays symbolic rather than inventing these numbers (uncertainty, 2026-10-23).
• No zero-knowledge component is included: an LP's own signature and PoolManager state verify the payout condition, so a ZK proof would not remove a remaining trusted party. The qualitative merit of a plan remains unprovable in the MVP (Inference, 2026-10-23).

13. Sources with read dates

All sources below were read 2026-10-23 unless a source's own publication/commit date is stated. Record = direct documentation, API, commit metadata or read-only chain state; Claim = a project's statement or an audit finding scoped to its candidate; Inference = this report's judgment, arithmetic or proposed parameter. Mutable APIs are pinned by observation times/blocks above. No downloaded project code was executed and no deployment was made.

• IMD docs (read 2026-10-23; published control-plane notes include 2026-10-10 and 2026-10-13), capabilities (read 2026-10-23), swarm (snapshot 2026-10-23 19:44:34 UTC), version (commit 962b3018e66f6ed5471ecf947a26101596c3fa61).
• Uniswap v4-core Hooks.sol, IHooks.sol, PoolManager.sol (commit 2026-10-23); Uniswap VolatilePair announcement (published 2026-10-14); official hook deployments (commit 2026-10-23).
• ERC-8004 official contracts/deployments (commit 2026-08-15); ERC-8004 spec (same pin).
• CLAUS live fee state (snapshot block 26,257,500 at 2026-10-23 19:44:42 UTC), CLAUS about (read 2026-10-23), IMD candidate host audit (read 2026-10-23; its scope is a candidate and not proof of current live owner delay), IMD fee-vault review (read 2026-10-23; separate project).
• BRO-AMM (commit 2026-10-15), Hindsight (2026-09-03), Tab (2026-10-18), Obelisk (2026-09-25), Veil (2026-10-15), Pines (2026-10-05), Even (2026-10-14), Tenet Agents (2026-10-05). Commit dates and SHA values came from GitHub API commit metadata read 2026-10-23.
• Read-only onchain address checks at Ethereum block 26,257,514 / 0xf06d44a68db62d994febf9363ec2200043502bdb0a55263a0ea5b408aef39501 and Robinhood block 96,193,417 / 0x7bde1d362117d4d9730582b4b74e4e17a7fbbdfb0f1e219f9fe1793f46b3184c, both 2026-10-23: Ethereum Intake, Robinhood Intake, CLAUS hook, CLAUS TradeExtensionHost. The CLAUS proxy-admin owner change was checked at blocks 26,197,787 and 26,197,788 on 2026-10-23. These are code/state checks, not verified-source claims.