# Claus arms: a bounded, bond-backed checker network **Research cut: 12 November 2026.** This is an independent desk-research view, not a statement by Claus or IMD. Tags mean: **[Read]** directly observed in a cited public source or public read-only chain/API response; **[Inference]** a conclusion from those observations; **[Proposal]** my recommended rule; **[Limit]** a measurement or evidence boundary; and **[Open question]** something only the founders can decide. Dollar figures are service prices and cost assumptions, not token-price forecasts or investment advice. ## Bottom line **[Proposal]** Do not turn the 300 houses into checking seats and do not make a permanent seat purchasable by burning CLAUS. Start a 90-day, capped *arm* pilot: an operator posts a returnable 100,000-CLAUS bond, passes a reproducible trial, completes independently matched control-check forms, and can lose a bounded portion of its bond only for a precisely disproven, objective claim. It is materially better than a rent-extracting NFT seat, but it is not trustless while one key can upgrade the CLAUS system; lock that key before taking third-party bonds. **[Inference]** The proposal is strongest as paid, evidence-bound verification, not as token economics: a second matching arm reduces ordinary error, but cannot prove two arms are independent or make subjective security opinions objectively slashable. This report therefore cuts subjective verdict slashing, unlimited concurrent liability, immediate burn of surplus, and the claim that “one controller, one voice” is enforceable from wallet addresses alone. ## 1. Corrected factual base * **[Read]** The official current CLAUS contract is `0x163015276f3244bC2CcB78b4D69cc8A4069EfFAC` on Robinhood Chain 4663; the site calls the Ethereum token retired for migration purposes. The site also says the token, hook and vault can be upgraded. [S1][S2] * **[Read]** The current official-pool project fee is 2%: 1% for CLAUS buyback/burn and 1% ETH to CLAUS DEV, not an undifferentiated 2% split. [S3] * **[Read]** The migration guide says 3,805 direct allocations were delivered 1:1. Separately, all 300 NFTs received 50,000 CLAUS each—15,000,000 CLAUS to 148 wallets, including staked NFTs attributed to their stakers. Thus “15m to 148 wallets” is correct only for the NFT payment; it is not the full migration-recipient count. [S4] * **[Read]** The guide says NFT migration is separate and earlier earned rewards retain their existing terms. It says legacy Ethereum tokens remain transferable, although post-snapshot activity creates no migration allocation. Therefore “the old token is retired” must not be read as “all Ethereum-side rights or transfers ceased.” [S2][S4] * **[Read]** Claus says Jobs have set prices, acceptance criteria and escrow on Ethereum. I found no public source establishing the exact five-person/four-vote IMD panel rule stated in the brief, so I treat that rule as unverified context rather than a fact. [S5] * **[Read]** At Robinhood block **112,496,202**, I called `totalSupply()` and `decimals()` through the public RPC: supply was **983,872,663.400486075752008619 CLAUS**, decimals 18. `owner()` on the token returned contract `0x979d…bdc46`; that contract’s `owner()` returned `0xbb145ca83272c3806d4ddc75ca1d5514789cf1c5`, whose code was `0x` at that head. The token owner is therefore a controller contract ultimately owned by an EOA at that snapshot—not literally a token owned directly by one plain wallet. The official site independently says the original developer wallet controls minting, fees and project-owned liquidity. [S2][S6] ## 2. A — seat choice and lessons from comparable systems **[Proposal]** Use an open, returnable work bond plus a trial, but call it an operator admission credential rather than a “seat.” An arm has no transferable economic right, no emissions and no guaranteed job flow. It earns only for accepted work; the bond exists to make a limited, falsifiable warranty credible. | Model | What it gets right | Failure mode | Verdict | |---|---|---|---| | NFT seat (current IMD-style) | Scarcity makes assignment and identity simple. IMD documents 2,000 NFTs and a share to seats connected at admission. [S7] | It turns past ownership/uptime into an entitlement; transferability separates operator skill from the right to be paid. | **[Proposal] Reject for Claus.** | | House-based seat | A visible house can be useful directory/social identity. Claus calls its 300 NFT houses a City feature. [S5] | Location/collectible ownership is unrelated to accuracy; 300 early owners become a gate. | **[Proposal] Keep houses as cosmetics only.** | | Open permit, no bond | Lowest barrier and best experimentation. | A wrong answer costs the customer but has no credible source of recovery; reputation is slow and easy to reset. | **[Proposal] Use only for unpriced sandbox trials.** | | Delegated staking | Holders can select specialists without operating themselves. | Delegation tends toward a few recognizable operators; published work on delegation explicitly identifies increased concentration on a small number of delegatees as a collective risk. [S8] | **[Proposal] Do not use for v1.** It reintroduces passive yield and voter apathy. | | Open bonded arm | Skill, available capital and a passed trial—not collectible ownership—admit the worker. | Capital can still exclude capable operators and volatile collateral can make liability irrational. | **[Proposal] Best fit, with the constraints below.** | **[Read]** These mechanisms are real, not merely theoretical. UMA’s Optimistic Oracle requires both asserter and disputer bonds and resolves an undisputed assertion as true after its liveness window; a dispute sends the result to its resolution path and the winner receives bonds less any configured fee. [S9] Kleros has operated since 2018 and says it has handled over a thousand cases; selection probability rises with stake, while incoherent/non-responsive jurors can lose locked stake and fees. [S10][S11] Optimism’s live fault-proof design similarly uses an interactive dispute game and an initial bond; its specification assumes an active honest challenger. [S12][S13] **[Inference]** Those examples support bonds and challenge windows, but not a claim that they remove concentration or apathy. Kleros’s own documentation says it cannot ensure jurors actually review evidence; it relies on incentives, and stake raises selection probability. [S10][S11] A tiny checker market has a sharper problem: monitoring a $250 form may cost more than the possible challenger award, so no one watches. The remedy is paid independent re-execution and a funded dispute reserve, not hoping that idle holders become vigilant. ## 3. B — bond, capacity, time and volatility **[Proposal]** Set the base bond at **100,000 CLAUS per arm** for the pilot. It is 0.0101659% of the observed supply, large enough to create a real exit cost yet not a permissioned fixed inventory. Use exactly this unit for admission and the E calculations below; do not represent it as a dollar promise. **[Proposal]** Classify a job before assignment: | Class | What the arm may attest | Bond capacity | Liability tail | Why | |---|---|---:|---:|---| | C1 control map | Enumerated owner, implementation, role, mint, pause, fee setter and liquidity-custody facts at a stated block | 1 active job per arm | 30 days after accepted verdict | Facts can be reproduced from ABI/code/storage and a pinned block. | | C2 frozen upgrade | C1 plus a diff from a supplied immutable implementation and a verified timelock | 1 active job, no C1 concurrently | 45 days | More evidence and more ways to omit a privilege. | | C3 watch | Change detection against an already accepted C1/C2 manifest | 2 watches, no C2 concurrently | 7 days after each alert/non-alert interval | The claimed fact is “a specified event/change did or did not occur,” not “the system is safe.” | **[Proposal]** Require the full bond to remain locked while capacity is used; a job may consume no more than one capacity slot. On exit, cancel unassigned work, finish the tail, then wait **45 days** before withdrawal. The 45 days is deliberately longer than the 14-day challenge window plus a 30-day target for an external decision; a dispute freezes the affected bond until final settlement. **[Inference]** A fixed CLAUS bond makes the operator’s real economic exposure move with CLAUS, so a competent operator can become under-secured or over-exposed without doing anything. No-emission offsets are (i) price jobs in dollars at quote time but settle CLAUS using a documented 30-minute median of two public venue quotes, (ii) pay arms promptly in that settled CLAUS, (iii) let an operator top up or reduce *uncommitted* excess only between jobs, and (iv) keep liability capped by slots. This reduces, but cannot eliminate, collateral volatility; it is why the pilot must be low-value and have a hard job-value ceiling. **[Proposal]** Do not create delegated CLAUS staking. A holder can hire or sponsor an operator off-protocol, but the registered arm alone supplies bond, trial identity, liability and payout address. That avoids passive security yield and the delegated-stake concentration problem. ## 4. C — a challenge game that can actually be adjudicated **[Proposal]** Make only an atomized statement slashable: for example, “at Robinhood block X, `owner()` returned A”; “implementation slot was B”; “role R included A”; “the proxy emitted Upgrade to B”; “timelock delay was D”; or “transfer T moved custody from A to B.” Each must include chain, block/hash, method/event/storage location, raw return/log, decoding version and source hash. A conclusion such as “safe,” “honest,” “rug-proof,” “well designed,” or an unbounded future prediction is non-slashable commentary. **[Proposal]** Use blind commit/reveal for the two arms: each commits a hash of the form plus evidence bundle, then reveals only after the counterpart commits. Exact byte equality is too brittle; matching means identical canonical fact keys and values, with a deterministic normalizer. A mismatch is *not* a slash—it becomes “unresolved” and is manually triaged or refunded. **[Proposal]** A challenger posts the lesser of **10% of the at-risk slash amount or $50 equivalent in CLAUS**, capped at $500, and supplies a conflicting canonical proof. The deposit prevents free spam but is not a pay-to-silence gate. One challenge per fact key is escrowed; identical later submissions join it instead of multiplying cost. Commit evidence first, reveal after the response deadline; reject duplicates and challenges without a machine-readable counterproof. This curbs griefing and makes collusive “challenge each other” attacks costly because a losing challenge pays the reserve. **[Proposal]** For one proven false C1/C2 fact, slash **20% of the arm’s bond**, not the whole bond. Distribute that slashed slice: **customer first, up to its paid fee plus documented re-check cost; then 40% of the slice to the successful challenger; 20% to the dispute reserve; and the balance burns.** The customer-first rule meets restitution without turning a $250 check into a windfall; the remaining numbers fund watchdogs and adjudication. A knowingly false attestation, forged evidence, or a second sustained error in the tail may slash the remaining bond only after external adjudication. An honest “unknown” or an explicitly scoped limitation is never slashable. **[Proposal]** Pay panel work from the dispute reserve at a published fixed schedule, never from the winning side’s discretionary reward. For claims about a contract IMD built, use a pre-committed external route: three randomly drawn, conflict-disclosed reviewers from a separately maintained roster, with a Kleros-style commit/reveal vote or a named independent arbitration provider. IMD may submit evidence but has no vote. If no outside route exists before launch, **do not sell checks of IMD-built contracts**. This is preferable to pretending an IMD panel can be impartial in its own case. **[Inference]** The structure follows the useful part of optimistic games—both sides put value at risk and the disputed claim is decided—but deliberately does not copy their computational scope. OP’s interactive games work because the disputed computation ultimately reduces to a VM instruction; a prose audit opinion does not. [S12] ## 5. D — fees and when burn is permissible **[Proposal]** “Arms first, reserve, coordination at cost, remainder burn” is sustainable only if the published price is above measured fully loaded cost. At launch, use the following *price floors*, review them after 30 completed jobs, and decline work below the floor: | Product | Customer price | Allocation before any burn | Reasoned cost basis | |---|---:|---|---| | C1 control map | **$250** | $85 each to two arms ($170); $30 trial/QA reserve; $30 disputes/refunds; $15 assignment/publishing; $5 retained surplus | Pays two independent runs rather than one answer and leaves a real error budget. | | C2 frozen upgrade | **$1,000** | $350 each to two arms; $150 QA/reserve; $100 dispute reserve; $100 coordination; $50 retained surplus | Diff/reproduction and timelock validation require substantially more review than field extraction. | | C3 watch | **$75/month** per manifest | $25 primary monitoring; $20 independent confirmation; $15 reserve; $10 coordination; $5 retained surplus | A cheap watch is only credible if at least two paths can detect/confirm a change. | **[Proposal]** Quote the dollar price, obtain a short-lived CLAUS settlement quote from the specified median, and escrow the resulting CLAUS amount before assignment. Do not expose the customer or arm to a hidden discretionary conversion rate. “Coordination at cost” must mean a published per-job ceiling and receipts/usage logs; otherwise it is a treasury cut under another name. **[Proposal]** Do not burn any operating surplus until the reserve holds the greater of **three months of trailing reserve obligations or 20 times the trailing 30-day dispute count times the mean C1 fee**. Until then, surplus stays segregated for refunds, planted defects, independent retests and panel fees. Thereafter burn only the monthly excess after liabilities are marked; never burn a customer’s refundable escrow. **[Inference]** The proposal’s immediate-burn rule is unsafe in a new market: early error/dispute frequency is unknown, and a burned dollar-equivalent cannot pay a late customer or arbiter. The existing Claus fee already funds compute/hosting/work through a treasury, so the claim that the new checker must have no working reserve does not follow from the current token design. [S3] ## 6. E — bond arithmetic and credibility of CLAUS collateral **[Read + calculation]** Source supply is the `totalSupply()` read above at Robinhood block 112,496,202, not a market-data site: 983,872,663.400486075752008619 CLAUS. At the recommended 100,000-CLAUS bond: | Registered arms | CLAUS bonded | Share of observed supply | |---:|---:|---:| | 10 | 1,000,000 | 0.1016% | | 100 | 10,000,000 | 1.0164% | | 1,000 | 100,000,000 | 10.1640% | **[Inference]** Those are accounting shares, not a forecast and not a claim about effective float: an arm may be one controller operating many addresses, a pooled-custody address may represent many people, and bonded tokens can be released after exit. They nevertheless show why 1,000 arms should not be an uncapped day-one target. **[Read]** Claus discloses that the Robinhood token and hook are upgradeable and that the original developer wallet controls minting, fees and project-owned liquidity. [S2] **[Inference]** An upgrade, mint, transfer restriction, fee change, or withdrawal of protocol-owned liquidity can change the usefulness and realizable value of a CLAUS bond without the arm being at fault; this undermines both deterrence and fair restitution. **[Proposal]** Before bonds are accepted, Claus must irrevocably lock the token implementation/upgrade authority, mint authority, transfer-pause/blacklist authority (or prove it does not exist), official-pool hook fee and fee recipient, and control of project-owned liquidity into a public timelock with a **minimum 30-day delay**, a visible queued-action registry, and an emergency action limited to pausing *new arm assignment*—not seizing, freezing or changing bonded balances. The ultimate EOA must not retain a bypass. An independent source/permission review must attest the exact calls and timelock configuration. If this cannot be done, use a non-CLAUS refundable bond or do not launch bonded checks. ## 7. F — demand, observable activity and prices **[Read]** The first credible buyers are projects about to launch or upgrade, launchpad users, funds/treasuries that must document control risk, and liquidity providers whose custody/fee path is opaque. Claus itself describes paid, checkable builder work including tests, ledger checks and watches, while IMD describes an audit flow that separates permission, math, economics and control-flow review. [S5][S14] **[Read + measurement]** At 12 November 2026, I paged the public `GET /launches?limit=500&before=` index, whose records expose a source GitHub repository and creation time. It returned **1,069** indexed launches from 21 August to 12 November. In the completed UTC days 1–11 November it recorded **15, 9, 2, 5, 0, 7, 8, 4, 4, 2, 7** launches respectively (63 total; 5.73/day); 12 November was partial (2 by 03:22 UTC). This is evidence of IMD-indexed launch/repository throughput, not proof that every source repository was independently built, nor a count of all on-chain token deployments. [S14] **[Read + limit]** I could not produce a defensible “new Robinhood tokens per day” count from a public explorer. The named Blockscout token page currently reports holders and transfers as N/A for CLAUS, and its public API returned a browser challenge in this research session; a standard public RPC does not enumerate all newly created ERC-20 contracts. [S15][S6] **[Inference]** A numerical chain-wide token/day assertion here would be invented. The minimum acceptable future measurement is: publish the explorer query or factory-event address, exact UTC window, block range, deduplication rule (proxy vs implementation, ERC-20 interface check), raw contract addresses and code hash. **[Proposal]** The price floors in section 5 fit the observed product hierarchy: $250 C1 is a narrow evidence form; $1,000 C2 pays for a frozen diff and two independent reviews; $75/month C3 is only offered for a prior accepted manifest. Do not launch a “cheap audit” tier or sell an assurance label. ## 8. G — the 300 NFT holders **[Inference]** The strongest objection is fairness: holders paid for or received houses in an ecosystem where NFT-linked rewards existed, and excluding them from a new cash-flow surface can feel like dilution of the social deal—even if it is not a contractual entitlement. The objection is strengthened by the fact that IMD’s own seat model currently gives connected seats a share of launches. [S7] **[Proposal]** The honest answer is: *a house is not a checker credential; paying old collectible ownership from checker fees would charge customers for someone else’s historical asset.* The NFT migration payment was already separate (15m CLAUS for 300 NFTs); preserving existing stated terms is appropriate, creating a new perpetual income right is not. [S4] **[Proposal]** Offer courtesy without income: a permanent City/ledger provenance page, a free read-only verification badge for a house’s linked project, and an invitation to the same public trial materials. Do not waive the arm bond, reserve jobs, distribute fee share, or give a governance veto. ## 9. H — concentration, method and the identity limit ### CLAUS on Robinhood Chain **[Read + measurement]** At block **112,496,202**, I replayed all ERC-20 `Transfer` logs for the official CLAUS contract from its first observed transfer at block **107,148,941** through the head via the public Robinhood RPC. There were 14,877 logs, 3,982 addresses with a positive replayed balance, and their sum exactly matched the `totalSupply()` result. Top-one/top-three/top-five/top-ten/top-twenty address shares were **9.8498%, 16.6942%, 21.6058%, 29.8921%, 40.5502%**. [S6] **[Read]** The top ten addresses and balances in CLAUS were: `0x8366…0951` 96,910,445.5432; `0x63b9…2cef` 40,309,609.9856; `0x44e4…d308` 27,029,617.2783; `0x1c57…0fb9` 24,924,923.4241; `0x9d42…b7ed` 23,399,613.1163; `0xabb0…3aa6` 18,374,326.8084; `0x27da…26e5` 17,184,444.8862; `0x1ab4…c308` 15,723,203.8632; `0x77b6…a44d` 15,667,440.1409; `0x1354…40a1` 14,577,301.6480 CLAUS. [S6] **[Limit]** Address concentration is not beneficial-owner concentration. Nine of those ten addresses had non-empty code at the head, so wallets, pools, smart accounts, escrow and controllers cannot be distinguished from balance alone; common control can make real concentration higher or lower. ### Claus NFTs and IMD seats on Ethereum **[Read + measurement]** Etherscan’s current holder table for Claus NFT `0x8039…95d9` reports 300 units. Its largest holder is the `ClausNftStaking` contract `0x4e5916428452891E1deb59d162c20266ED08CAD4` with **185 (61.6667%)**, followed by `0xc421…271B` with **21 (7.0%)**; top ten visible holder addresses hold **232/300 = 77.3333%**. The dominant address is custody, so this is an address-level upper bound on holder concentration, not a count of the beneficiaries—consistent with Claus’s statement that staked NFTs are attributed to stakers for the migration. [S16][S4] **[Read + measurement]** IMD identifies the seat collection as `0x0000ec93127baa929e58e97dd0095a2bfb38ec1d`, capped at 2,000. At API timestamp **2026-11-12T11:30:47.991Z**, its public chain-read owner endpoint returned 2,000 seats at 789 owner addresses. I counted the returned `owners` array: the largest ten addresses held **385/2,000 = 19.25%**—`0xb7a8…e8ff` 77, `0xb641…1d72` 54, `0x8027…aca2` 47, `0xe3de…d80c` 34, then three addresses with 30, two with 28 and one with 27. [S14][S17] **[Limit]** The IMD result is a current API chain read, not a pinned Ethereum block; the Claus NFT result includes a staking custodian; neither reveals common beneficial control. The earlier Etherscan search snapshot of 593 holders is stale and should not be substituted for this contemporaneous owner-endpoint count. [S18] **[Proposal]** “One controller, one voice” cannot be enforced from fresh wallet addresses. Do not claim it can. The cheapest workable *deterrent* for the pilot is one registered operator identity per payout address, a device-bound passkey, a trial video/liveness check with conflict declaration, and random rechecks; it is not proof of unique human control. Enforce the economically relevant rule instead: one registered arm may fill one side of a job, and assignments reject shared payout address, device key, employer declaration, evidence source, or declared common control. Pay no voting reward for additional wallets. For higher-value work, a founder must choose whether privacy-preserving personhood/KYC is acceptable; no technical shortcut supplies it. ## 10. What I would cut, and an executable pilot **[Proposal] Cut now:** permanent-seat burn; any NFT/house fee share; unlimited arms from day one; “two matching answers equals truth”; full-bond automatic slashing; subjective audit-opinion slashing; immediate surplus burn; delegated staking; IMD judging its own contracts; and the one-controller claim. **[Proposal] Build only this pilot:** 20 maximum arms; C1 only; 30 jobs maximum; 100,000 CLAUS each; 20 planted fixtures with public answer schema but hidden variants; require 16/20 correct, zero critical role/upgrade omission, and a clean rerun; one open C1 per arm; 14-day challenge window; 30-day tail; the published $250 floor; monthly public ledger of assignments, canonical facts, mismatch/refund rates, reserve balance, challenges and controller conflicts. Stop automatically after 90 days or on two sustained false facts, then publish the raw outcomes before considering C2 or C3. **[Read]** ERC-8004 is compatible with this limited approach: its stated trust models include validation via stake-secured re-execution and generic validation hooks, while payments are explicitly orthogonal. It is an identity/reputation/validation registry concept, not a reason to make an NFT into a paid seat. [S19] ## 11. Founder-only open questions 1. **[Open question]** Will Claus lock every relevant token, hook, mint, pause/blacklist, fee-recipient and project-liquidity control as specified before accepting bonds? Which exact contracts, functions and timelock signers are in scope? 2. **[Open question]** Who funds the initial reserve and external-arbiter roster, and may customer refunds exceed a single job’s paid fee? 3. **[Open question]** What privacy/identity trade-off will founders authorize for operator uniqueness, knowing wallet-only uniqueness is unenforceable? 4. **[Open question]** Which external arbitration organization or conflict-disclosed reviewer roster is authorized for IMD-built contracts, and who appoints/replaces it? 5. **[Open question]** Is the current ETH Jobs escrow to remain separate, or will a new Robinhood settlement contract be funded and audited? The public pages establish the former exists but do not specify this migration decision. [S5] 6. **[Open question]** Are prior NFT terms limited to the stated historical rewards/migration treatment, or is there a published contractual promise of future work-fee participation? I found no such public promise in the sources reviewed. ## 12. I — five sentences for a tired CLAUS holder 1. **[Proposal]** A Claus house stays your house, but it is not a lifelong right to take a cut from other people’s security checks. 2. **[Proposal]** Anyone who can pass the same hard trial and lock the same returnable CLAUS bond can become an arm, so useful work—not an old NFT—gets paid. 3. **[Proposal]** Two independent arms must show the same on-chain facts before a customer is charged, and a proved false fact pays the customer and challenger before anything burns. 4. **[Proposal]** No arm earns for sitting still, and no new fee stream is created for the 300 houses. 5. **[Proposal]** This only starts after the people controlling CLAUS can no longer quietly change the collateral rules underneath the workers. ## Sources and reproducibility notes * **[S1]** [Claus home / official Robinhood address](https://claus.si/). * **[S2]** [Claus contracts and control disclosure](https://claus.si/Contract). * **[S3]** [Claus Project Funding / current pool fee split](https://claus.si/read/Hooks/Funding). * **[S4]** [Claus migration guide](https://claus.si/docs/migration). * **[S5]** [Claus priorities and Jobs description](https://claus.si/Priorities). * **[S6]** [Robinhood Chain public RPC](https://rpc.mainnet.chain.robinhood.com) and [Robinscan’s chain/RPC description](https://www.robinscan.com/). Reproduction: `eth_blockNumber`, ERC-20 `totalSupply()` (`0x18160ddd`), `decimals()` (`0x313ce567`), `owner()` (`0x8da5cb5b`), `eth_getCode`, EIP-1967 implementation slot, and `eth_getLogs` filtered to the ERC-20 Transfer topic; replay credits `to` and debits `from`, excluding zero-address debits/credits. This is public read-only RPC evidence captured at the stated head. * **[S7]** [IMD documentation: 2,000 seats and seat/work shares](https://imd.fun/docs/). * **[S8]** [How to Rationally Select Your Delegatee in PoS](https://arxiv.org/abs/2310.08895). * **[S9]** [UMA Optimistic Oracle V3 implementation](https://github.com/UMAprotocol/protocol/blob/master/packages/core/contracts/optimistic-oracle-v3/implementation/OptimisticOracleV3.sol). * **[S10]** [Kleros FAQ](https://docs.kleros.io/kleros-faq). * **[S11]** [Kleros juror tutorial](https://docs.kleros.io/products/court/kleros-juror-tutorial). * **[S12]** [OP Stack fault-proof specification](https://specs.optimism.io/fault-proof/). * **[S13]** [OP Stack bond incentives](https://specs.optimism.io/fault-proof/stage-one/bond-incentives.html). * **[S14]** [IMD API reference: launches, audits, contracts and seat endpoints](https://imd.fun/docs/api/). The paged public endpoint used for the launch count is [api.imd.fun/launches](https://api.imd.fun/launches?limit=500). * **[S15]** [Robinhood Blockscout CLAUS page](https://robinhoodchain.blockscout.com/token/0x163015276f3244bc2ccb78b4d69cc8a4069effac). * **[S16]** [Etherscan Claus-NFT holder table](https://etherscan.io/token/generic-tokenholders2?a=0x80396c7131159eB92E7E84839e51C9887A7A95d9&s=300&i=0&sid=cb342bc6f91dbe3615ee4fc0f2f4e63f&p=1) and [staking-contract page](https://etherscan.io/address/0x4e5916428452891E1deb59d162c20266ED08CAD4). * **[S17]** [IMD public seat-owner endpoint](https://api.imd.fun/seats/owners), and the collection address in [IMD API documentation](https://imd.fun/docs/api/). * **[S18]** [Etherscan identity.md collection page](https://etherscan.io/token/0x0000ec93127baa929e58e97dd0095a2bfb38ec1d). * **[S19]** [ERC-8004: Trustless Agents](https://eips.ethereum.org/EIPS/eip-8004).