# Claus beside IMD: pay for independent work before paying for scarcity Researcher 4 · Economics and incentives · Public-source desk research · 12 November 2026 **[Proposal] My verdict:** pursue a paid upgrade-review pilot, but change the economic contract before turning houses into workers. Sell a bounded, independently checked result; pay its operators in money they can spend on inference; retain a reserve before burning; and make interoperability a purchase order rather than a reciprocal subsidy. Three hundred property NFTs are neither a demonstrated capacity constraint nor evidence of three hundred willing operators. **Reading key:** **[Read]** means a statement found in the linked public source, not independently certified operation. **[Inference]** means my deduction, including calculations and research limits. **[Proposal]** means my recommendation, including every suggested number, threshold and future mechanism. A label governs its paragraph, list item or table cell; mixed claims are labeled separately. Prices in examples are hypothetical dollars, not market quotations. Nothing below implies either founder has agreed. **[Read — research record]** I read public Claus pages and documentation, IMD documentation and `skill.md`, the public Claus repository, ERC-8004, x402 documentation and published Anthropic research. The Claus repository revision inspected was `b1852dda3878a8b7c4ecf4ab8e3d41afa6147b9f`; its pinned documents are linked below. IMD's public capabilities response was also read without authentication. No account, private endpoint, transaction, paid request or worker was used. The other researchers' views were not read. **[Inference — limits]** This establishes documented terms and exposes inconsistencies; it does not establish demand, current deposit admission, completed cross-swarm trades, active seats, treasury solvency or operator profitability. Some public Claus reads were unavailable, so I used accessible linked documentation and pinned public source, identifying that distinction. I did not obtain Huang's original post and therefore do not verify its date, numerical claims or the proposition that riders are the scarce input. That proposition remains a hypothesis to test through repeat customers. ## Correct the premises first | v0.1 premise | Evidence and correction | Economic consequence | | --- | --- | --- | | Claus has “no working seat, no coordinator.” | **[Read]** Claus documents bounded isolated JavaScript workers and a hosted scheduler. These have no network or credentials and are not continuous servers. Native agents separately have funded workspaces and pause/resume controls. [Worker guide][C12], [agent launch guide][C8]. | **[Inference]** A holder-operated review fleet is still an additional product, but existing execution and coordination must not be described as absent. Worker counts cannot be inferred from houses or public profiles. | | Claus has “no independent acceptance or onchain record.” | **[Read]** Jobs supports a reviewer selected before claim, evidence-bound reports and conflict checks. Reports are advisory; the requester still accepts. Paid Jobs separately commits delivery and settlement onchain. Coin changes already receive Claus review. [Jobs][C5], [coin-review criteria][C7]. | **[Inference]** The missing economic layer is enforceable, adequately funded independent verification with clear authority—not the invention of reviews or receipts. | | A bridge to IMD is wholly new. | **[Read]** The documented dispute path buys one IMD oracle opinion from five panel members, requires four to agree exactly, then requires separate Claus approval. Its initial subsidy is capped at eight cases; admission and completion are conditional. [Paid deals][C6]. | **[Inference]** Start with this documented precedent, without presenting it as proof of a currently functioning general bridge or independent demand. | | Inference is paid by the treasury rather than users. | **[Read]** NFT sponsorship is a separately funded allowance up to $5/NFT/day, with automatic top-up off. It does not run an always-on agent. Native launchers separately deposit 10 USDC operating credit on Base; agent trading fees can replenish operating budgets. [NFT guide][C2], [sponsorship conditions][C3], [agent launch guide][C8]. | **[Inference]** There are several payers and funding paths. Retiring universal subsidies does not require abolishing useful, bounded sponsorship or user-funded hosted agents. | | Claus has no posted price; its services are generally x402/USDG. | **[Read]** Pinned customer-service documentation describes fixed-price project reports paid by direct Base USDC transfer and explicitly says this is not an x402 seller endpoint. Pinned service documentation describes another, separately admitted direct-prepayment/testnet path. Work profiles can quote ETH or CLAUS. [Customer services][C10], [services][C11], [work profiles][C9]. | **[Inference]** “No pricing” is too broad. I did not verify a current uniform upgrade-review tariff or a general production USDG seller path. A sample price ceiling in documentation is not a live price. | | The existing 2% fee can finance the new service. | **[Read]** The official CLAUS Robinhood pool charges 2%: 1% buyback/burn and 1% operating treasury. Native agent coins have a distinct starting 1.5% trading fee and a Claus share with a 0.5%-of-output floor, allocated to CLAUS buybacks. [Hooks][C1], [agent launch guide][C8]. | **[Inference]** Neither trading volume nor fee percentages demonstrate enough cash to fund a review fleet. Do not combine these different revenue streams or assume token launches create recurring customer revenue. | | “The rider owns it” follows automatically from launch. | **[Read]** For native Claus coins, the agent proposes changes and Claus controls reviewed execution. The review criteria distinguish execution-service safeguards from the powers of the root administrator. [Coin-review criteria][C7]. | **[Inference]** Workspace ownership, project-token holdings and contract administration are different rights. The joint offer must specify which rights the rider actually receives. | | IMD's request price is a suitable economic template. | **[Read]** Public capabilities lists 0.5 IMD for a job and 600-second quotes; groups are priced per question. IMD's onchain guide says payments are forwarded to a payee/treasury and there are no refunds. Its overview instead says there is no treasury and network earnings are removed from supply. [Live capabilities][I6], [onchain guide][I4], [overview][I1]. | **[Inference]** Payment admission, downstream burn and worker compensation are different flows. The public wording does not reconcile them. Do not claim a verified 100%-burn request split or infer cost recovery from the 0.5 number. | | IMD acceptance proves a review's truth. | **[Read]** IMD's API reference says text-output verification checks structure alone; it distinguishes this from independent adversarial review. [Review documentation][I7]. **[Read]** ERC-8004 is a draft identity/reputation/validation interface, with payments and validator incentives outside its scope and explicit Sybil limitations. [ERC-8004][E1]. | **[Inference]** Onchain acceptance can be a faithful record of a weak check. Publish the actual checks and economic responsibility; do not price a registry entry as an assurance guarantee. | **[Read]** IMD documents 2,000 seats on holders' own machines/plans, and launch rewards of 8% to seats connected at admission plus 2% to accepted-work wallets. Its detailed launch guide also specifies 1.25% trading fees, with 1% to the payer and 0.25% to the network, and excludes contracts-only deployments from token allocations. [IMD overview][I1], [launch guide][I5]. **[Inference]** v0.1 therefore omits a material existing launch cash flow and overstates how much new coordination Claus needs. The important distinction is between owning a scarce asset, supplying qualified work, buying a result and accepting liability for it. ## Five changes, ranked ### 1. Replace the universal flat token price with a bounded service and a cost-covering tariff **[Proposal — change]** Publish one standard upgrade-review package first: a pinned implementation and diff, specified chain/state snapshot, a defined dependency boundary, two separately operated runs, executable checks where applicable, findings and an explicit inconclusive outcome. Larger changes receive a new quote. Neither a favorable verdict nor unlimited rework is included. Hosting and continuous monitoring are separately priced renewals. **[Read — evidence]** Anthropic reports that its research multi-agent system consumed about 15 times the tokens of ordinary chats and needed sufficiently valuable tasks to justify that cost. This is evidence from that system, not a Claus cost estimate. [Anthropic engineering report][A1]. **[Inference — reason]** Fixed CLAUS units against variable dollar inference costs create two failure modes: falling token value makes requests loss-making, while rising token value prices customers out. An unlimited brief also lets customers select the most expensive work at the average price. Copying IMD's admission price would copy neither its costs nor its launch-based rewards. **[Proposal — price rule]** Account in dollars and settle the Claus service in CLAUS. Let `P` exclude separately itemized downstream swarm purchases, network gas and conversion costs; let `Cw` include operator inference, tools, machine costs, expected retries and minimum compensation; let `Co` include coordination, hosting and dispute provision. Quote only if `0.75 × P ≥ Cw` and `0.20 × P ≥ Co` after allowing for conversion risk. Use observed pilot costs, including failed jobs, to set `P`; do not publish a fabricated “market price.” Freeze the exact payable CLAUS amount for five minutes using an executable conversion quote with size and slippage limits. If reliable conversion is unavailable, decline the order. | Destination | Proposed mature-service share of `P` | Obligation | | --- | ---: | --- | | Performing operators | **[Proposal] 75%** | **[Proposal]** Initially 45% lead run, 30% independent review; pay against completed rubric-defined work, including an adverse conclusion. | | Operating treasury | **[Proposal] 20%** | **[Proposal]** Coordination, hosting, dispute reserve and any rider reward; no additional rider levy. Publish actual costs and residual margin. | | CLAUS burn | **[Proposal] 5%** | **[Proposal]** Only after settlement and required reserves are funded. During the pilot use 0% burn and 25% treasury. | **[Proposal — illustration]** At hypothetical `P = $100`, a mature allocation is $75/$20/$5. If fully loaded worker costs are $60 and platform costs/provision $15, worker surplus is $15 and platform residual is $5 before any rider reward. If worker costs rise to $90, the worker constraint alone requires at least $120; retaining a $100 tariff would hide a subsidy. These numbers demonstrate the rule; they are not recommended security-audit prices. **[Proposal — inference payer]** The customer ultimately pays inference inside the package. The operator pays its provider and accepts work only if its compensation covers that cost. Permit a genuinely authorized own-plan worker, metered API access, or a user-funded hosted runtime. Do not require every NFT owner to buy a subscription. Keep NFT-sponsored inference separate: count its consumed value as a disclosed subsidy during experiments, never as free commercial capacity. Do not promise renewal or use a transferred NFT to reset an allowance. **[Inference — holder participation]** Some technically capable holders might run profitable workers; public documentation cannot establish that most would. A willingness to hold a house is not demonstrated willingness to supply uptime, provider quota and incident response. IMD's worker documentation itself cautions that earnings are not guaranteed. [IMD skill][I2]. **[Inference — subsidy bound]** If all 300 NFTs had funded periods and consumed their full $5 daily allowance, sponsorship would cost up to $1,500/day or $45,000 over 30 days. Funding that solely from the official pool's 1% treasury slice would require $150,000/day of trading before other operating costs. These are conditional upper-bound calculations, not observed usage or an unconditional entitlement. Customer-service margins should sustain commercial reviews even when trading revenue falls. [NFT allowance][C2], [treasury fee][C1]. **[Proposal — buyer protection and cost]** Take a small, disclosed admission fee only for actual screening, then reserve the execution budget under milestone terms. Unstarted work is refundable; completed adverse review is payable; disputed quality goes through the agreed process. Hold refund reserves in the owed assets and operating reserves in assets matching provider bills; do not count unsold project tokens at headline market capitalization. Do not spend liabilities on burn. Burn activates only after funded obligations plus 90 days of measured fixed operating costs are covered. This costs immediate burn narrative, requires pricing/accounting and dispute handling, and may reveal that commercially useful reviews cost much more than token holders expect. Existing x402 payment transport needs an application-level reserve/refund arrangement; x402 itself does not supply the service contract. [x402 introduction][X1]. ### 2. Separate the 300 houses from operating labor; use delegation before issuing another scarce asset **[Proposal — change]** Preserve the 300 property identities and their existing rights. Let each house authorize one active, revocable operator permit, with a specified service role and expiry, while its holder keeps custody. The operator supplies the machine/provider account and receives the work fee; any holder/operator split is agreed before dispatch. Never share a personal model login with the delegate. Attribute each result to both the house and the operator's current control interval. **[Read — evidence]** The NFT guide distinguishes the fixed 300-house neighborhood from ordinary projects, and says ownership transfer does not transfer private project history or keys. [NFT guide][C2]. **[Read]** IMD's worker distribution uses the connected machine's authenticated runtime and quota. [Worker repository][I8]. **[Inference — are 300 enough?]** There is no economically meaningful minimum NFT count for a swarm. Illustratively, capacity is `N × a × h / (k × t)`, with `a` the qualified available fraction, `h` usable hours/day, `k` independent operators/request and `t` hours/operator/request. With 300 houses, `a=0.20`, `h=4`, `k=2`, `t=2`, capacity is 60 reviews/day before queue headroom, coordination and correlated failures. Thirty pilot seats at `a=0.60` yield 18/day under the same assumptions; at a 70% utilization ceiling that is 12.6/day. All inputs are assumptions, not observed capacity. A hundred houses controlled by one operator do not supply a hundred independent opinions. **[Proposal — best second class]** The first additional class should be renewable, nontransferable **delegated operator permits tied to existing houses**, not burn-to-mint NFTs. If independent demand later exceeds the qualified 300-house fleet, admit qualified overflow contractors on expiring permits paid solely for incremental work. They receive neither house property rights nor the house-specific capacity allocation proposed in change 5. Use successful blind benchmark work and attributable control to qualify; admission should not depend on buying or destroying CLAUS. **[Inference — dilution tradeoff]** Burning tokens can demonstrate willingness to pay, not ability to review upgrades or remain independent. Reversible bonds are more useful than irreversible burns when a narrow, provable offense needs recourse. No design can guarantee unchanged expected house earnings while opening an overlapping labor market: additional capable workers can lower rents even without increasing NFT supply. **[Proposal — protection, trigger and cost]** Protect the 300's enumerated property and governance rights and any expressly contracted house allocation; promise no fixed investment return. Add overflow only after qualified utilization exceeds 70% and deadline misses exceed 5% for two measured weekly cohorts, with customer demand exceeding existing capacity. Require a funded demand forecast and publish displacement effects. This costs exclusivity rents and operator-management work. It also means a house can remain a house while a specialist performs its work. Do not rewrite existing sponsorship terms without the holders' required authorization. ### 3. Replace reciprocal hiring obligations with pre-funded, isolated purchase orders **[Proposal — change]** Keep each swarm's native settlement token. A rider's single checkout can fund a routing account that purchases exact downstream amounts; any stablecoin used at checkout is a payment convenience, not a shared swarm token. Quote the bill as `IMD purchases + Claus purchases + conversion/gas + disclosed routing fee`. Apply the Claus 75/20/5 split only to its own service revenue, never the gross pass-through payment for IMD work. **[Proposal — example]** If an order needs two IMD requests and one Claus review, reserve exactly 1 IMD plus the quoted CLAUS amount and disclosed conversion/gas before dispatch. This arithmetic uses IMD's documented per-request price, not a claimed exchange rate. If the buyer starts with CLAUS, the buyer's funded order buys the IMD; if IMD hires Claus, its funded order buys CLAUS. No bilateral IOU, guaranteed exchange rate or treasury-backed cross-token credit is created. [IMD price documentation][I3]. **[Inference — value movement]** Hiring creates revenue for the supplier and cost for the purchaser. Token demand at checkout can be offset when workers sell tokens to pay providers. A burn reduces units; it does not prove that the hiring swarm or its holders earned a positive dollar return. Repeated A→B→A orders can increase both networks' reported volume while consuming their reserves. **[Proposal — drain controls]** Give every order a total cost ceiling, maximum depth of two subcontracting hops, explicit permitted suppliers and no repeated service cycle. A supplier may not use its buyer's treasury or auto-replenish a balance. Separate new customer deposits, earned revenue and finite promotional grants. Pause dispatch on stale prices, excessive price impact, exhausted reserves or unknown prior payment state. Set token purchase limits, per-counterparty daily loss limits and an aggregate subsidy ceiling before the pilot. Never interpret an RPC timeout as permission to pay again. **[Read — integration constraints]** IMD's HTTP payment path in the live response uses Ethereum while its onchain intake includes Robinhood Chain. Claus's documented optional Jobs funding requires Ethereum wallets; its published token-deal record is explicitly historical and names the older Ethereum CLAUS asset. These documents do not prove a Robinhood CLAUS escrow deployment. [IMD capabilities][I6], [Jobs][C5], [dated escrow record][C13]. **[Read]** x402 can express arbitrary EVM chains/tokens, but production requires an actual compatible settlement path. [x402 token/network support][X2]. **[Proposal — implementation boundary and cost]** Reuse the documented IMD-oracle/Claus-approval path where currently admitted; verify exact chain, asset, escrow version, availability and settlement receipt before promising more. Preserve no-refund IMD admission terms in the customer's quote; refund only the unspent Claus/routing portion unless a separately funded guarantee covers more. This costs working capital, exchange fees and some conversion friction. A failed downstream service can strand a partially completed order; disclose that failure allocation rather than imply atomic execution across swarms. **[Proposal — third swarm]** Keep the six-question swarm card, but make its answers machine-readable and versioned, including chain/asset, price expiry, scope limits, acceptance/check profile, refund and dispute rules, available capacity and receipt format. A third swarm joins through the same purchase-order rules and budgets. It receives no claim on either existing treasury and no obligation to buy back work. ### 4. Pay reviewers for honest conclusions and riders for durable, attributable outcomes **[Proposal — change]** Separate three records: performed work, buyer settlement and subsequent outcome. A job needs a globally unique parent ID across swarms, pinned inputs/output, charged amounts, identified operators/reviewers, conflicts, check results and final payment status. Preserve rejected attempts, refunds and unresolved states. Record the maturity observation at 30 and 90 days: for a review this means its reproducible findings and subsequent material escapes within scope, not a website merely remaining reachable. **[Read — evidence]** Claus already says paid review compensation is for the review, including failed or inconclusive conclusions. Its normal report does not stop escrow deadlines. [Jobs review terms][C5]. **[Read]** ERC-8004 warns that identities and public feedback do not eliminate Sybil manipulation. [ERC-8004][E1]. **[Inference — reason]** Paying by requests ordered, spend, favorable acceptance or number of findings makes circular purchases, rubber stamps and low-value reports profitable. A different swarm name does not make the second review independent when owners, models or evidence overlap. A rider should not earn a commission simply by spending its own or an affiliated wallet's money. **[Proposal — reviewer incentives]** Assign reviewers independently of the implementer, check known controllers, obtain conflict disclosures and keep first-pass findings blind. Where feasible vary tools/model families as well as seats. Pay the fixed verification fee for a complete evidenced report regardless of conclusion; reward reproduced material findings only from a separately capped bounty pool. Do not slash for disagreement, declining unsafe work, an inconclusive result or an undetected defect alone. A refundable bond may cover only demonstrable fraud, duplicate billing or an explicit missed commitment under a contestable rule. For all custody/permission-changing upgrades, require a qualified external-swarm review before execution, even if the request fee is small; if independence or funding is absent, do not execute. | Rider benefit | Proposed rule | What prevents volume farming | | --- | --- | --- | | Priority | **[Proposal]** Earn through matured, on-time, undisputed outcomes; cap the priority lane at 10% of capacity. Urgency and paid service commitments still govern. | **[Proposal]** No priority from raw spend; cap influence per known operator/customer group and retain ordinary/newcomer access. | | Credit | **[Proposal]** Start with earned, nontransferable future-service rebates, not unsecured lending. Rebates cannot pay gas or externally purchased services. | **[Proposal]** No leverage or treasury claim; issue only from realized margin, expire unused rebates after 90 days and reserve their fulfillment cost. | | Fee share | **[Proposal]** A disclosed customer-acquisition/coordination reward, up to 10% of the platform's realized contribution margin attributable to unaffiliated external customers, payable after 90 days. | **[Proposal]** Exclude grants, self-orders, related counterparties, pass-through spend, refunded jobs and duplicate milestones; no payment when contribution margin is nonpositive. | **[Proposal — payout example]** For the $100 example in change 1, $5 of platform residual permits at most $0.50 of rider reward, paid from the treasury share, leaving $4.50. A project token's paper valuation, swap turnover or repeated cross-swarm receipts cannot enlarge this reward. Where counterparty independence cannot be established, publish the record but withhold the cash reward. **[Proposal — cost and portability]** Use one shared receipt schema with verifiable mirrored references; avoid two inconsistent balances or a transferable reputation token. Keep historical NFT records, but start a new controller interval for earning priority/credit after transfer. Qualification and payout require attribution and delayed observations, which cost privacy, administrative effort and immediate incentive strength. Small cohorts are easy to game; do not claim that public wallet history alone proves independent customers. Keep these rewards optional until measurement costs are known. ### 5. Give the company a renewable service budget and negotiate one launch contribution pool **[Proposal — change]** The rider controls the company assets specified by the launch terms. The company owes each swarm the price of services actually ordered: build/change milestones to IMD, hosting and monitoring renewals to Claus, and independent review commissions to whichever qualified swarm supplies them. Pre-fund 90 days of expected operation and review on a rolling budget. Publish renewal, incident, upgrade and shutdown obligations. No company owes a perpetual tax merely for being launched, and no swarm owes indefinite free support. **[Inference — reason]** A launch allocation is speculative compensation for initial contribution and risk. It cannot substitute for paying next month's inference and hosting. Automatically giving both networks the full IMD-style allocation would increase total extraction from 10% to 20% before proving twice the value. Giving idle connected seats most of the tokens also weakens the incentive to perform the expensive work. **[Proposal — future joint-launch policy]** Negotiate a **single total 10% contribution pool**, agreed before work. An illustrative allocation is below; it needs a new expressly approved distribution policy, not creative reinterpretation of today's IMD entitlements. | Recipient/purpose | Proposed percentage of total project-token supply | Earning condition | | --- | ---: | --- | | IMD build contributors | **[Proposal] 5%** | **[Proposal]** Accepted build milestones, attributable work, vesting over 90 days. | | IMD reserved capacity | **[Proposal] 1%** | **[Proposal]** Measured availability committed for this project, not an admission-time connection snapshot. | | Claus independent reviewers | **[Proposal] 1%** | **[Proposal]** Completed independent review work; verdict-neutral compensation. | | Claus operating contributors | **[Proposal] 2%** | **[Proposal]** Earned over the first 90 days against the agreed operating service, with unearned units returned to the requester under published terms. | | Claus house capacity | **[Proposal] 1%** | **[Proposal]** Within the fixed 300-house class, pro rata to qualified reserved capacity; no new house claims for overflow contractors. | | Total | **[Proposal] 10%** | **[Proposal]** IMD 6%, Claus 4%; 8% attributable work/operation and 2% capacity. | **[Proposal — existing-system fallback]** Until IMD agrees and implements a different policy, honor its documented 8% connected-seat plus 2% accepted-work distribution. Pay Claus in cash from the company's explicit budget. If the requester additionally offers project tokens to Claus, disclose the exact transfer from the requester's allocation and resulting ownership before launch; do not call it a redistribution of IMD's 10%. Never add another automatic 10%, and never assume a token-free company has a launch-token allocation. [IMD launch terms][I5]. **[Read — ownership trap]** IMD's onchain guide distinguishes contract owner, token remainder recipient and fee recipient. The paying address receives the 1% trading-fee entitlement by default; continuing the project remains tied to that payer. [Onchain guide][I4]. **[Proposal — company control and cost]** Show those separate rights in the rider's checkout. If a router pays, do not silently let it keep the trading-fee entitlement or make the rider dependent on it for continuation; arrange the supported fee handoff and explicitly address the payer-continuation restriction before sale. Set an authorized incident decision-maker, a finite remediation reserve and an exit package covering source, data, hosting transfer and revocation. Stop promising “no staff”: autonomy is an operating goal, with a named accountable owner and paid escalation. This costs vesting/distribution work, delayed rewards, less passive seat income and a less spectacular launch promise. Lifetime revenue shares, if any, must be separately priced, time-bounded and buyout-capable; they are not implied by hosting. ## What I would cut - **[Proposal]** Cut burn-to-mint seats. They tax entry without buying competence, and place expansion decisions at the mercy of token price. - **[Proposal]** Cut universal flat CLAUS pricing, automatic burn before liabilities are reserved, and mandatory holder-funded inference. Each hides a different subsidy or excludes useful specialists. - **[Proposal]** Cut compulsory paid Claus calls at the end of every IMD workflow. Offer them where hosting, operations or a genuinely additional review is useful; do not manufacture reciprocal revenue. - **[Proposal]** Cut rider rewards for ordering, spending or merely shipping once; cut unsecured reputation credit in the pilot. - **[Proposal]** Cut automatic duplicate launch allocations, perpetual operating obligations bought with illiquid tokens, and the unconditional “no staff” promise. - **[Proposal]** Cut calendar-driven expansion to all 300 and then a third swarm. Expansion should follow external repeat purchases, positive unit economics and sufficient independent operators. ## What I would keep exactly as is - **[Proposal]** Keep adjacency rather than copying IMD, with IMD primarily building and Claus primarily housing, operating and reviewing. - **[Proposal]** Keep “no shared token; each swarm is paid in its own.” Dollar accounting and a convenience checkout do not alter that settlement principle. - **[Proposal]** Keep upgrade review as the first coordinated Claus job type, and the initial 30-seat pilot. - **[Proposal]** Keep phase 1's three Claus-ordered IMD jobs and three IMD builds published in the City as integration exercises. Label related-party spending and sponsorship; those six exercises do not prove product-market fit. - **[Proposal]** Keep the swarm card's six questions verbatim: who are your seats, what do you sell, what does it cost, how do I pay, what do I get back, who checked it. - **[Proposal]** Keep one brief and one payment as the rider-facing goal, and public, mutually readable work records. The price and failure allocation must be explicit behind that simplicity. ## Open questions only the founders can answer **[Proposal — questions, not inferred facts]** Public research cannot supply the following internal commitments or policy choices: 1. What exact property, governance, sponsorship and economic rights have already been promised to the 300 houses, and which changes require holder consent? Will Claus authorize operator delegation without conveying those rights? 2. What are actual inference, hosting, coordination and remediation costs, unrestricted liquid runway and outstanding liabilities? How much finite cash can the founders commit to the pilot, including worst-case failed orders? 3. Will IMD reconcile the overview's “no treasury/burn” language with its documented payment forwarding and launch-fee recipient? Which flows actually fund operating costs and workers, and which are burned, under the current policy? 4. Which Claus payment/review paths do the operators currently authorize for production on Robinhood Chain, and what continuity commitments will they make? The old Ethereum CLAUS escrow is not permission to treat the new token as interchangeable. 5. Are both teams willing to change future launch distributions, preserve existing entitlements and support explicit fee/continuation rights for a rider whose router pays? Who can authorize these changes? 6. Who accepts commercial liability for a missed upgrade defect or prolonged outage, within what cap, and who may stop execution or arbitrate a dispute involving either founding team? 7. What external repeat-customer revenue, margin and review-quality thresholds will justify expansion? Will either team accept publishing a failed pilot and stopping subsidy rather than counting reciprocal orders as success? 8. Who controls the rider-facing service and its attributable customer relationships? What ownership/control evidence will the founders require for cash rewards and credit, accepting the associated privacy tradeoff? ## Revised design, one-page version **[Proposal — purpose]** Claus becomes IMD's adjacent specialist for upgrade review, hosting and operation. The joint product is a purchased outcome with a known budget and accountable owner. No shared token. Each swarm receives its native token; a router can offer one brief and one funded checkout while showing scope, third-party charges and partial-failure terms. **[Proposal — first service]** Sell a bounded upgrade-review package against a pinned change and state snapshot. Require two separately controlled runs, reproducible evidence, explicit limitations and an independent review for custody/permission-changing upgrades. A complete adverse or inconclusive review is paid work. Preserve the owner's decision and existing contract controls; a receipt does not certify safety. **[Proposal — supply]** Retain the 300 house identities and existing rights. Begin with 30 qualified seats. A house may delegate one expiring operating permit; operators bring authorized inference or a funded hosted runtime. The customer covers inference through the service price. Keep NFT sponsorship separately budgeted and disclosed. Add no burn-minted seat. Admit overflow contractors only when measured demand and utilization justify them, without new house-specific rights. **[Proposal — price]** Quote each bounded package in dollar terms, payable in an exact short-lived CLAUS amount. Price from fully loaded worker and platform costs. Split the pilot's own service revenue 75% performing operators and 25% treasury; after liabilities and 90 days of fixed costs are reserved, use 75%/20%/5% to operators/treasury/burn. Exclude pass-through purchases and gas. Return unspent budget; state the nonrefundable admission and completed-work portions before payment. **[Proposal — federation]** Begin with existing intakes and the documented IMD-opinion/Claus-approval precedent, checking current admission, chain and token. Reserve downstream payments before work, isolate customer funds, cap conversions and prohibit circular subcontracting. Unknown payments retain their original identifiers. Neither side guarantees the other's token price or extends automatic credit. Keep the six-question swarm card; version its answers and include commercial terms and verification evidence so a third swarm uses the same rules. **[Proposal — riders]** Mirror verifiable records of orders, attempts, checks, settlement and 30/90-day outcomes. Deduplicate cross-swarm milestones. Reward matured outcomes with a priority lane capped at 10%, funded future-service rebates and at most 10% of realized platform contribution margin from unaffiliated customers. Exclude self-orders, subsidies, pass-through fees and refunds. No unsecured lending; no cash reward where independence is unverified. **[Proposal — company]** The company pays IMD for builds and changes and Claus for renewable hosting, operations and review. Budget 90 days ahead, identify an incident owner and offer an exit package. Propose one future 10% launch contribution pool, weighted 8% toward work/operation and 2% toward committed capacity, with earning/vesting conditions. Until both teams authorize that policy, retain IMD's existing 8%+2% and pay Claus explicitly in cash. Disclose owner, remainder, fee and continuation rights separately. No perpetual unpaid support or automatic second allocation. **[Proposal — gates]** Phase 0 resolves policy authority and budgets. Phase 1 performs the original three jobs and three City publications as labeled integration exercises. Phase 2 recruits 30 operators and measures cost, independence, completion, defects, reserve use and external repeat purchase. Proceed only after two measured weekly cohorts show positive contribution margin excluding grants, at least 90% on-time completion, at most 5% unresolved service disputes, funded reserves and at least three unaffiliated customers buying again. Phase 3 enables mirrored records and fully funded cross-review. Phase 4 expands capacity only on the utilization trigger in change 2, then admits a third swarm and one pre-funded company. These are proposed decision rules, not claims that either network currently meets them. ## Sources and reproducibility **[Read — source register]** All links below were read on 12 November 2026. Live web documentation can change. Pinned repository links identify source documentation, not proof of deployed availability. Public capabilities was retrieved by an unauthenticated GET; no quote or purchase was created. These notes and the research limitations above define the evidential scope of this report. [I1]: https://imd.fun/docs/ "IMD overview and seat economics" [I2]: https://imd.fun/skill.md "IMD public hiring and worker skill" [I3]: https://imd.fun/skill/pay.md "IMD HTTP payment guide" [I4]: https://imd.fun/skill/onchain.md "IMD onchain payment, refunds and payer rights" [I5]: https://imd.fun/skill/jobs.md "IMD launch and job terms" [I6]: https://api.imd.fun/requests/capabilities "IMD live public capabilities" [I7]: https://imd.fun/docs/api/#reviews "IMD review verification limits" [I8]: https://github.com/Identity-md/worker "IMD public worker distribution" [C1]: https://claus.si/Hooks "Claus official pool economics" [C2]: https://claus.si/docs/nft-inference "Claus NFT inference and houses" [C3]: https://claus.si/docs/nft-inference-terms "Claus funded-period sponsorship conditions" [C5]: https://claus.si/docs/jobs "Claus Jobs, payment and independent review" [C6]: https://claus.si/docs/paid-deals "Claus escrow and IMD dispute-review path" [C7]: https://claus.si/docs/coin-reviews "Claus native coin-change criteria" [C8]: https://claus.si/guides/agent-launch "Claus native launch, operating credit and fee economics" [C9]: https://claus.si/docs/agent-economy "Claus work profiles and reviewer compensation" [C10]: https://raw.githubusercontent.com/contractclaus/claus-network/b1852dda3878a8b7c4ecf4ab8e3d41afa6147b9f/docs/customer-services.md "Pinned Claus customer-service payment documentation" [C11]: https://raw.githubusercontent.com/contractclaus/claus-network/b1852dda3878a8b7c4ecf4ab8e3d41afa6147b9f/docs/services.md "Pinned Claus service protocol documentation" [C12]: https://claus.si/docs/workers "Claus bounded workers and hosted scheduling" [C13]: https://claus.si/docs/token-deal-deployment.json "Historical Claus Ethereum token-escrow deployment record" [E1]: https://eips.ethereum.org/EIPS/eip-8004 "ERC-8004 draft: interfaces, incentives and security limits" [X1]: https://docs.x402.org/introduction "x402 payment protocol introduction" [X2]: https://docs.x402.org/core-concepts/network-and-token-support "x402 protocol versus production network/token support" [A1]: https://www.anthropic.com/research/multi-agent-research-system "Anthropic: How we built our multi-agent research system, 13 June 2025"