# Claus as a swarm adjacent to IMD: what I would change in the economics (view 2 of 5) Researcher 2, working alone. Lens: economics and incentives. Written 2026-11-12. **Labels used throughout** - **[S]** read from a cited public source (the source number is in brackets, e.g. [S3]) - **[C]** read from Ethereum mainnet state with a read-only `eth_call` (block 26,397,865 / 0x192cca9) - **[I]** my inference from sources - **[P]** my own proposal - **[?]** unverified or uncertain **Limits.** This is desk research on public pages only, fetched 2026-11-12 around 09:10 UTC. I did not log in, sign, or call any private or write endpoint. Claus's pages change often and say so themselves [S1]. Every figure below is a snapshot. I could not find Matt Huang's post of 12 Nov 2026 on "swarm riders" with a web search [S20], so I treat the core thesis as the proposer's premise and not as a sourced fact **[?]**. --- ## 0. Corrections first: facts v0.1 gets wrong or out of date | v0.1 says | What the sources say | Why it matters economically | |---|---|---| | Claus has "a 2% pool fee, half to buy and burn CLAUS, half to compute and hosting" | The Robinhood official pool takes a 2% project fee, **all of it to liquidity**: "No current treasury share", and a keeper adds liquidity in batches [S1][S2]. The Ethereum pool takes **2.3%, all accruing as ETH to one fixed recipient** (ProjectRevenue, "CLAUS DEV") [S1]. New NFT, LP-reward and buyback allocations are paused [S1][S3]. The current priority is "build Robinhood liquidity with the complete 2% project fee" [S1]. | **[I]** No fee stream exists today that could fund burns, seats or compute. Any seat pay has to come from request revenue, not from the pool. v0.1's request split cannot assume a burn that already exists. | | "300 NFT houses" that can become seats on Robinhood Chain | The 300 NFTs are ERC-721 on **Ethereum (chain 1)**. Each "represents 50,000 deposited $CLAUS" of the Ethereum token, and the owner can burn one to redeem them [S3][S4]. "NFTs, backing and stakes remain on Ethereum" [S1]. All 300 exist: `totalSupply` = 300 [C], "available": "0" [S5]. In the snapshot 185 are staked in the vault [S5]. Counting the beneficiary of staked items, there are **about 147 distinct holders**. The largest holds 22, and holders of 10 or more hold 91 of the 300 [S5, my count]. | **[I]** The seat set is small and concentrated. Each seat also carries ~50,000 CLAUS of redeemable backing. Holders have a cheap exit: they can redeem, which removes the seat. Seat identity, ERC-8004 registration and pay probably live on Ethereum, while requests are paid on Robinhood Chain. | | "Sponsored inference up to 5 dollars a day" (implied: enough to run work) | Up to $5 per NFT per day, auto top-up off, model `claude-haiku-5.5` only [S6]. Live policy: 1,024 max output tokens, 10,000 max input characters, 6 requests a minute, concurrency 1. **`supports300Independent: false`** [S7]. It runs through Surplus Intelligence, with no zero-retention guarantee, and "does not host an always-running agent" [S8]. | **[I]** This allowance cannot do upgrade review: a 1k-token Haiku reply on a 10k-character input is far below what IMD requires for contract work [S11]. The platform also does not attest that it can serve all 300 at once. It is an onboarding perk, not a worker substrate. | | "x402 services in USDG" | Claus's own paid service (Contract Control) settles in **USDC on Base** [S9]. USDG over x402 appears in a third-party City project ("undertow") [S2]. | **[I]** The x402 rail exists, but the stablecoin and chain are Base/USDC for Claus's first-party service. Pricing "in CLAUS over x402" would be a new payment path, not an extension of an existing one. | | "395 API operations" | The current OpenAPI has 426 operations on 377 paths [S10, my count]. | Minor. The number grows, so the design should not quote it. | | IMD: "flat 0.5 IMD a request" with seats paid somehow | 0.5 IMD a request, whatever the steps or panel size [S11][S12]. Onchain, it is "**forwarded whole to the treasury in the same call**" [S12]. The same docs also say "Nobody takes a fee. There is no treasury. What the network earns is removed from supply" [S12] **[?] these two statements conflict**. Seats are paid **only through launches**: 8% of each launched token to connected seats, 2% to wallets with accepted work, via a Merkle distributor. They also get an onchain ERC-8004 reputation review per accepted job [S12]. | **[I]** IMD seats earn nothing per request. Their pay is a lottery on launches. That model does not carry over to a review job, which launches nothing. | | Riders/seats "earn priority" (as at IMD) | IMD's router ranks by tools, load, failures and turn. "**Past wins count for nothing**". Only contract work has probation [S11]. | **[I]** IMD deliberately does *not* give a good record more work. v0.1's "record earns priority" is new on both sides, not something to copy. | | Claus has "no independent acceptance or onchain record" | Mostly right. On Claus Jobs only the requester accepts. Acceptance "is not a Claus security audit" and reviews are "advisory; they cannot … pause escrow deadlines" [S13][S14]. A claim lasts one hour [S13]. CLAUS deal terms are read with `chainId: 1` [S13], so **[I]** escrow appears to be on Ethereum. Claus does write its own Journal releases onchain (`claus-journal/1`), which is a publisher record, not an acceptance record [S1]. | Confirms the gap. | | Upgrade review as first job (implicit: it protects holders) | The Robinhood token, hook and liquidity management are **owner-upgradeable by a single admin wallet with extra mint authority, no onchain mint cap or timelock**. The 1-hour notice "is not enforced by the contracts" [S1]. | **[I]** A review that cannot delay an upgrade is worth little to anyone, so nobody will pay much for it. See change 3. | Also confirmed: IMD seat NFTs `totalSupply` = 2,000 [C]. They bind to ERC-8004 agent ids, and reviews are written to the ERC-8004 Reputation Registry [S12]. ERC-8004 defines Identity, Reputation and Validation registries. Its trust models are "tiered, with security proportional to value at risk", including "validation via stake-secured re-execution" [S15]. --- ## 1. Answers to the lens questions ### 1.1 Are 300 seats enough, and what second seat class? - **[S]** IMD has 2,000 seats [C] and gives every step to one machine except research and audits, which go to a panel [S11]. Claus has 300 seats and ~147 holders [S5]. - **[I]** For one job type, upgrade review, 300 is enough. Upgrades to a token, hook or vault are rare events. A 30-seat pilot, about 20% of holders, is a realistic ceiling for who will run a worker on their own plan. - **[I]** 300 is not enough for a general swarm that also builds. That is why v0.1 is right that IMD builds. - **[I]** The real constraint is not seat count but **operators**: holders who will keep a capable model online. The thesis itself says the scarce input is the people who can command swarms. - **[P] Second class: delegated operators first, bonded seats second. No burn-for-seat.** - *Delegated operator:* a house holder who will not run a worker names an operator, who registers an ERC-8004 agent tied to that house. Pay splits between holder and operator. The default I would propose is 70% operator / 30% holder, changeable by the holder. This adds capacity without adding a single seat, so the 300 are not diluted. - *Bonded seat (phase 4 only):* anyone locks a fixed CLAUS bond, for example 50,000 to mirror the NFT backing [S4], for a non-transferable work seat. It gets job offers only after the 300 and delegated seats are busy. It gets **no share of launch allocations reserved for houses**, and leaving returns the bond. A bond lets ERC-8004's "stake-secured re-execution" model put value at risk [S15]. Rejected work can lose a seat its bond's priority, not its tokens. Slashing would need a dispute process nobody has built. - *Why not burning CLAUS for a seat [I]:* a burn is a one-way purchase. It sells work access to whoever has capital and creates pressure to recover the burn through volume, which is the behaviour the rider record is meant to discourage. A bond can be returned and is the same price for everyone. ### 1.2 What should a Claus request cost, and how should it split? - **[S] Facts.** IMD charges a flat 0.5 IMD, sends it whole to a treasury, and pays seats only from launches [S12]. Claus Jobs prices are per-task agreements in ETH or CLAUS, and "ETH and CLAUS prices are never compared as if they were equal" [S14]. Claus's x402 service quotes a fixed USDC price per order and refunds on a problem reported within seven days [S9]. - **[I]** A flat price works at IMD because seats are paid from launches. A review job launches nothing, so **the request fee is the only thing that pays the seats**. Most of it has to go to them, or holders will not spend their own plan on it. - **[P] The price:** one posted price per upgrade-review request. - Quoted in USD terms, set by Claus and published in its swarm card, and **paid in CLAUS** at a 30-minute pool TWAP at quote time, so a volatile CLAUS does not reprice the work. - As a working number for the pilot, I would propose an amount above a capable model's marginal cost for two independent runs plus a reviewer. I cannot source that cost **[?]**. The founders should set it from the pilot's observed spend (open question Q3). - **[P] The split:** | Share | Recipient | When | |---|---|---| | 70% | Seats with accepted work: builder run, independent second run, reviewers, equal per accepted role | Paid on acceptance | | 10% | Burn | — | | 20% | Treasury, for operating the coordinator, hosting and disputes | — | - **[P]** If no seat's work is accepted, the 70% does not go back to the payer as cash. It goes to the treasury as a credit the requester can spend on a retry. This keeps IMD's no-refund simplicity [S12] without letting failed work be a free option. - **[I] Why the burn is only 10%:** burn rewards passive holders. The scarce input is operators and riders. A large burn share pays the people who are not doing the work. The pool fee already funds liquidity [S1]. Changing that is a separate decision (Q2). ### 1.3 Who pays for inference? Will holders run a worker? - **[S]** IMD workers run on the holder's own Claude Code or Codex plan, and IMD tells them "a worker is not guaranteed earnings" [S16]. Claus's sponsored allowance is Haiku-only, small and not attested for all 300 [S6][S7]. Claus's project inference was switched off for new submissions on 21 Oct 2026 [S17]. - **[P]** The **seat holder pays inference on their own plan, as at IMD**. The treasury's $5/day sponsorship becomes a bounded onboarding grant, for example a house's first 30 days. It is used for house publishing and light tasks, never for review runs. - **[I] Will holders run one? A few dozen, if paid per accepted job.** - For: a Claus seat already ties up ~50,000 CLAUS of backing [S4]. Holders are concentrated: one wallet holds 22 seats [S5]. A single operator with many houses has a reason to run workers if per-job pay exists, as IMD's community runs "several NFTs on one server" [S12]. - Against: there is no pay today. The NFT reward share is 0 and paused [S3]. - Hence change 1. Without per-job pay, I expect almost none **[I]**. ### 1.4 With no shared token, how does value move? Can one side drain the other? - **[P] Rule: pay in the payee's token at the payee's posted price, per call, with no netting and no credit lines.** - When Claus orders an IMD job, the Claus side (requester or treasury) acquires IMD and pays 0.5 IMD into IMD's intake on Robinhood Chain. The intake accepts IMD on chain 4663 [S11][S12]. - When IMD hires Claus, IMD pays Claus's posted CLAUS price. - **[I]** Each cross-hire is a buy of the payee's token and a sell of the payer's, so value moves through the two markets, not through a bridge or a shared ledger. - **[I] How one side can drain the other:** 1. **Price mismatch.** IMD's price is flat whatever the number of steps [S11]. A Claus coordinator could resell IMD's large multi-step workflows inside a higher Claus price, so IMD seats carry the cost and Claus keeps the margin. This is acceptable only if IMD chooses it. It is IMD's pricing, not the bridge, that creates the opening. 2. **Mandatory reciprocal review.** v0.1's "work above a set value gets its second review from the other swarm" makes one swarm work for the other's customers. If the payee is unpaid or under-paid, it is drained. 3. **Treasury-funded cross-orders.** Phase 1 has Claus order three IMD jobs. If the Claus treasury pays and sells CLAUS on a thin Robinhood pool to get IMD, CLAUS holders bear the price impact. - **[P] Guards:** - The requester pays the cross-review price on top, never the other swarm. - Each swarm publishes a per-epoch cap on cross-orders it funds from its treasury. - Each cross-call settles on its own chain and in its own token, and is recorded in both swarms' records. ### 1.5 How should the rider record pay out? - **[S]** IMD records accepted work per agent onchain and gives no routing priority for wins [S11][S12]. Claus portfolios link accepted tasks to receipts, and "application acceptance, independent counterparties and payment are separate facts" [S14]. - **[P] The record counts outcomes only.** A rider's score counts jobs that were (a) paid at the posted price, (b) accepted by the independent check, not by the rider, and (c) **still live** at 30 and 90 days: site resolves, contract unchanged or upgraded only after a review. - Counts are weighted by distinct counterparties. Jobs where the rider also owns the accepting seat or the deliverable's operator wallet are excluded. - Volume alone earns nothing, and the score decays over 180 days. - **[P] Payout, in this order:** 1. **Priority:** a higher score gives a place earlier in the queue. This costs the swarms nothing. 2. **Credit:** a rebate, in the payee's own token, of up to 10% of the rider's past spend on jobs that are still live at 90 days. It is paid by the swarm that did the work. 3. **Fee share:** only from revenue of swarm-native companies the rider owns (see 1.6), never from request fees. A request-fee share would pay riders for ordering more, which is volume. ### 1.6 What does a swarm-native company owe each swarm? How do launch allocations work with two swarms? - **[S]** IMD's launch share: 8% equally to seats connected at admission, 2% equally to wallets with accepted work [S12]. - **[P] Keep the total swarm share at 10%, not 20%.** Split by role: | Share | Recipients | When | |---|---|---| | 2% | Wallets with accepted work, across both swarms | Immediately, as today | | 5% | IMD connected seats | At admission, as today but smaller | | 3% | Claus houses that actually host, operate and review the company | Vesting over 12 months. Unvested amounts go to the replacement operator if Claus stops operating it. | **[I]** Doubling the swarm share to 20% because two swarms touched it would tax every launch for the coordination and would not scale to a third swarm. - **[P] Over its life, the company owes each swarm only posted-price payments for services it uses.** That means Claus hosting, review before every upgrade, and IMD builds of next versions (`job.continue` [S11]). It owes nothing else: no perpetual royalty. Staying is optional, and leaving costs the unvested 3%. - **[P]** As an ownership rule, the rider gets owner roles, as IMD already does: the `owner` gets contract owner roles [S11]. The company's contracts should carry a timelock at least as long as the review service level (see change 3). --- ## 2. The five changes to v0.1, ranked **1. Pay seats per accepted job from the request price. Do not rely on a pool split or launches.** [P] - *Reason:* the pool fee funds liquidity only [S1], NFT rewards are paused [S3], and IMD's per-request revenue goes to a treasury while seats are paid from launches [S12]. A review job launches nothing. Without per-job pay there is no reason to run a worker [I]. - *Change:* the posted CLAUS price splits 70 / 10 / 20 between seats (on acceptance), burn and treasury. Failed work becomes retry credit. - *Cost:* a higher price per request than a burn-heavy design. The treasury also needs a settlement contract on Robinhood Chain that pays ERC-8004-identified seats whose NFTs sit on Ethereum, which means either a cross-chain ownership read or a holder-registered payout address (Q5). **2. The seat holder pays inference. Sponsorship becomes an onboarding grant only.** [P] - *Reason:* the sponsorship is Haiku-only, 1,024 output tokens, 6 per minute and not attested for 300 seats [S6][S7]. It cannot do contract review. IMD's own-plan model is proven [S16]. - *Cost:* fewer seats online, because only holders who pay for a capable plan show up [I]. Claus loses "free AI for holders" as a marketing line. **3. Make upgrade review binding: an onchain timelock on the Robinhood token, hook and vault upgrades, at least the review service level.** [P] - *Reason:* today a single admin can upgrade and mint with no timelock, and the 1-hour notice is not enforced [S1]. A review that cannot delay an upgrade has no price anyone will pay. With a timelock, the review becomes the one thing standing between a proposed upgrade and holders' funds [I]. - *Cost:* Claus gives up instant upgrades and fixes. It needs an emergency path, for example pause-only with no upgrade. The first coordinated job becomes reviewing that timelock contract itself, built by IMD. **4. Cross-swarm settlement rule: the payee's token at its posted price, per call, no netting. Cross-review is opt-in and paid by the requester, with per-epoch treasury caps.** [P] - *Reason:* this is what stops either side draining the other (1.4), and it works unchanged for a third swarm with a third token [I]. - *Cost:* riders pay two tokens for one brief. The "one payment" promise of the rider layer then needs a router that swaps, which is added work and a slippage risk on thin pools. A mandatory cross-review above a value threshold is dropped. **5. The rider record pays on shipped-and-still-live outcomes. Priority first, credit second, fee share only from owned companies.** [P] - *Reason:* v0.1 gives priority, credit and fee share without saying what earns them. A fee share on requests pays volume. IMD already shows a router can give priority without a "wins" loop [S11]. - *Cost:* a 30/90-day liveness check is a recurring oracle job, which IMD can sell at 0.5 IMD per question [S11]. Riders are rewarded late, and early riders see little benefit until phase 3. --- ## 3. What I would cut - **"Half to buy and burn CLAUS, half to compute and hosting"** as a description of today. It is wrong [S1]. Any burn should come from the request split, and stay small. [P] - **Treasury-paid inference for seat work.** [P] - **Mandatory second review from the other swarm above a set value.** Make it opt-in and paid by the requester. [P] - **"One brief, one payment" before phase 4.** It needs a two-token router. In phases 1 to 3, one brief with two payments is honest and enough. [P] - **Fee share from request revenue for riders.** [P] - **Phase 4's bundle** of "all 300 seats + a decision on a second seat class + the first swarm-native company + opening the card". It is four bets in one phase. Delegated operators move into phase 2, so the pilot can reach 30 operators. Bonded seats wait until review demand exceeds the 300 plus delegates. [P] ## 4. What I would keep exactly as is - **No shared token.** Each swarm is paid in its own. It is the only rule that scales to a third swarm. [P] - **The division of labour:** IMD builds, Claus houses, operates and reviews. IMD already sells builds, launches, workflows and audits [S11]. Claus already hosts and runs an apps and jobs platform [S2][S13]. [I] - **Upgrade review as the first coordinated job.** It is the job Claus's own holders most need, given single-admin upgradeability [S1]. Change 3 makes it worth paying for. [I] - **ERC-8004 for seats and the acceptance record.** IMD already writes verdicts to the ERC-8004 Reputation Registry [S12], and the standard scales trust with value at risk [S15]. [S/I] - **The swarm card's six questions.** One clarification [P]: "what does it cost" means price, token, chain and the payment rail (x402 or intake), as IMD's capabilities endpoint publishes [S18]. - **Phase 0, a yes or no, and phase 1, hiring each other with what exists.** It is cheap and uses live interfaces: IMD's intake on chain 4663 [S11] and Claus static hosting and services [S2][S9]. [I] ## 5. Open questions only the founders can answer 1. **Claus:** will you put a binding onchain timelock on the Robinhood token, hook, liquidity management and mint authority, and how long can it be? 2. **Claus:** will the Robinhood 2% stay all-liquidity, or will any part fund seats, burn or compute? Will the paused NFT reward share ever resume? 3. **Claus:** what posted price per upgrade review (USD terms, paid in CLAUS) are you willing to charge, given that 70% must cover two capable model runs plus reviewers? 4. **IMD:** which statement is current: "forwarded whole to the treasury" or "there is no treasury"? Where do 0.5 IMD request fees end up [S12]? 5. **Both:** where are seat payouts settled, given that Claus NFTs are on Ethereum and requests are on Robinhood Chain? Is a holder-registered payout address acceptable? 6. **IMD:** will you accept a 5% / 3% / 2% split of the 10% swarm share for jointly operated companies, or keep 8% / 2% and let Claus earn only from posted service prices? 7. **IMD:** do you want Claus-routed demand at the flat 0.5 IMD for multi-step workflows, or a separate price for orders arriving through another swarm's coordinator? 8. **Claus:** the default holder/operator split for delegated seats, and whether a bonded seat class is acceptable to current holders at all. 9. **Both:** who maintains the rider record, which chain it lives on, and who pays for the 30/90-day liveness checks? 10. **Proposer:** a link to the Huang post. I could not verify it **[?]** [S20]. --- ## 6. Revised one-page design (v0.2, my proposal [P] unless marked) **Goal.** Claus becomes the second swarm beside IMD. It does review and operations, not a copy of IMD's builds. The link between them is a set of settlement rules any third swarm can adopt. **Facts it rests on [S].** Claus has 300 house NFTs on Ethereum, each backed by 50,000 CLAUS, held by ~147 holders [S3][S5][C]. The CLAUS token is on Robinhood Chain and owner-upgradeable with no timelock [S1]. The Robinhood pool fee is 2%, all to liquidity [S1]. IMD has 2,000 ERC-8004 seats on their holders' own plans and charges 0.5 IMD a request. IMD seats are paid via launches: 8% to connected seats, 2% to accepted work [S12]. **Seats.** - The 300 houses register ERC-8004 agents and run on the holder's own plan. - Delegated operators can run a house's agent for a split. - Bonded seats (refundable CLAUS bond, no launch share) are considered only once review demand exceeds supply. - Sponsored inference remains an onboarding grant only. **First product: upgrade review.** - Claus adds a timelock on its own upgrades, reviewed first, built by IMD. - Each review is one posted price, in USD terms, paid in CLAUS over x402 or an intake. - It runs as an independent second run plus reviewers. The verdict is written to each seat's ERC-8004 record. - Split: 70% to accepted seats, 10% burn, 20% treasury. Failed work becomes retry credit. **Between swarms.** - Pay the payee's token at its posted price, per call, with no netting and no credit lines. - Cross-review is opt-in and paid by the requester. - Each swarm caps treasury-funded cross-orders per epoch. - Claus orders IMD work through IMD's intake on Robinhood Chain. IMD workflows can end in a paid Claus hosting or review call. **Riders.** - The record counts paid, independently accepted, still-live (30/90 days) outcomes, weighted by distinct counterparties. Self-dealing is excluded, and the score decays over 180 days. - It pays priority first, then credit (up to 10% of spend on still-live jobs, in the payee's token). - Fee share comes only from companies the rider owns. **Swarm-native companies.** - IMD launches them, the rider owns them, and Claus hosts, operates and reviews them. - The swarm share stays 10%: 2% accepted work (both swarms), 5% IMD connected seats, 3% Claus operators vesting over 12 months to whoever keeps operating it. - After launch, the company pays only posted prices for services it uses. Its contracts carry a timelock at least as long as the review service level. **Swarm card.** Six questions: seats, what you sell, cost (price, token, chain, rail), how to pay, what you get back, who checked it. A third swarm joins by publishing one and accepting the settlement rule. **Phases.** - **0:** yes or no on the timelock and on per-job seat pay. - **1:** three IMD jobs ordered by Claus, three IMD builds hosted in the City. Requesters pay; treasuries are capped. - **2:** 30 operators (houses plus delegates) do upgrade review at the posted price. The first job is the timelock. - **3:** opt-in cross-review, the swarm card and the rider record go live. - **4:** first swarm-native company. Decision on bonded seats. Card opened to a third swarm. --- ## Sources All fetched 2026-11-12 unless noted. Public, read-only. - [S1] Claus, llms.txt: https://claus.si/llms.txt (migration and fees, token control, pool fees, journal records, NFT access) - [S2] Claus, product reference: https://claus.si/product-reference.json (Claus Jobs description, City projects including "undertow" USDG x402) - [S3] Claus, project JSON: https://claus.si/about.json (`nfts`: ERC-721, chainId 1, capacity 300, backing 50,000, `sharePpm: 0`, `newRewardsPaused: true`) - [S4] Claus, collection configuration: https://claus.si/nft-config.json (collection 0x8039…95d9 on chain 1, `backingTokens` 50000) - [S5] Claus, live collection JSON: https://claus.si/nfts.json (block 26,397,865; `totalStaked` 185; 300 items; holder counts computed by me from `owner`/`beneficiary`) - [S6] Claus, NFT inference and agent accounts: https://claus.si/docs/nft-inference - [S7] Claus, NFT capabilities (public read): https://claus.si/__network/v1/nft/capabilities - [S8] Claus, NFT inference conditions v2026-10-23-v1: https://claus.si/docs/nft-inference-terms - [S9] Claus, Services (Contract Control, USDC on Base, refunds): https://claus.si/services - [S10] Claus, OpenAPI: https://claus.si/__network/openapi.json (operation count by me) - [S11] IMD, skill.md: https://imd.fun/skill.md and worker reference https://imd.fun/skill/worker.md ("Who goes next", probation, model requirements) - [S12] IMD, docs as Markdown: https://imd.fun/docs/llms.txt ("What a seat earns", "The entire ecosystem runs on $IMD", intake price "forwarded whole to the treasury", swarm share via Merkle distributor, ERC-8004 adapter and registries, no refunds) - [S13] Claus, Jobs guide: https://claus.si/docs/jobs (one-hour claim, requester-only acceptance, ETH/CLAUS agreements, `deal_terms({chainId: 1, asset: 'CLAUS'})`) - [S14] Claus, Work in Claus Network: https://claus.si/docs/agent-economy (work profiles, quotes in ETH or CLAUS, advisory reviews, portfolios) - [S15] ERC-8004, Trustless Agents: https://eips.ethereum.org/EIPS/eip-8004 - [S16] IMD, worker section of skill.md: https://imd.fun/skill.md ("A worker is not guaranteed earnings", own Claude Code or Codex plan) - [S17] Claus, Project inference: https://claus.si/docs/inference ("New project inference requests are disabled as of 21 October 2026") - [S18] IMD, live capabilities: https://api.imd.fun/requests/capabilities - [S19] x402: https://www.x402.org (open standard for HTTP-native payments; background only) - [S20] Web search for "Matt Huang swarm riders swarm native companies", 2026-11-12: no matching post found. Results included https://paradigm.xyz/team/matthuang and https://www.antoinebuteau.com/lessons-from-matt-huang/, neither of which uses the terms. - [C] Ethereum mainnet `totalSupply()` read-only calls at block 0x192cca9: Claus collection 0x80396c7131159eB92E7E84839e51C9887A7A95d9 → 300; IMD seat NFTs 0x0000ec93127baa929e58e97dd0095a2bfb38ec1d → 2,000. Public endpoint for the chain, as named by IMD: https://eth.drpc.org