# Claus arms: a CLAUS bond as the seat. View 1 of 5 Desk research, 12 Nov 2026. Independent view; nothing here is agreed by Claus or IMD. No price predictions, no investment language. **Marks used on every statement** - **[S]** read from a public source in this run (source given). - **[R]** recalled from published writing I have read before; the link is to the primary document but I did **not** re-fetch it in this run. Treat numbers as needing a check. - **[I]** inferred by me from [S] facts. - **[P]** my own proposal. - **[U]** unknown or not verified; I say why. **Sources read in this run** - S1 swarmcard v0.3 public page: https://swarmcard.imd.site/ - S2 Claus home: https://claus.si/ - S3 Claus hooks page: https://claus.si/Hooks - S4 Claus migration doc: https://claus.si/docs/migration - S5 IMD docs: https://imd.fun/docs/ - S6 IMD skill file: https://imd.fun/skill.md - S7 IMD explorer token list: https://explorer.imd.fun/tokens - S8 Robinhood Chain public RPC `https://rpc.mainnet.chain.robinhood.com` (chain id 4663), read-only `eth_call` / `eth_getStorageAt` / `eth_getCode` at block 112,496,406 - S9 Ethereum public RPC `https://ethereum-rpc.publicnode.com`, read-only `eth_call` at block 26,398,605 - S10 GitHub public API, org `Identity-md`: https://api.github.com/orgs/Identity-md/repos - S11 Blockscout docs naming the Robinhood Chain explorer: https://docs.blockscout.com/robinhood-api.md --- ## 1. Corrected facts | # | Context says | What I found | Mark | |---|---|---|---| | 1 | CLAUS is on Robinhood Chain at `0x1630…fFAC` | Confirmed. Chain id 4663. `symbol()` returns CLAUS, 18 decimals. | [S] S2, S8 | | 2 | Pool fee 2%, half buyback and burn, half treasury | Confirmed: "1% buys and burns CLAUS and 1% accrues in ETH for CLAUS DEV". The legacy Ethereum pool charges 2.3% to the same treasury. | [S] S3 | | 3 | 300 NFTs paid 50,000 CLAUS each, 15 million to 148 wallets | Confirmed, "from the existing reserve"; staked NFTs were assigned to their stakers. The NFT contracts remain on Ethereum. | [S] S4, S3 | | 4 | "The token is owned by a contract owned by one plain wallet" | True but **incomplete**. `owner()` = `0x979D5810D32292358639f25294472f01c95bDc46` (a contract, about 2.2 kB of code); its `owner()` = `0xBb145cA83272c3806D4dDc75cA1D5514789cF1C5`, which has no code (plain wallet). **But the token is an upgradeable proxy** (about 740 bytes of code, delegatecall pattern) whose admin, read from the EIP-1967 admin slot and also hard-coded in the bytecode, is a **third address `0xcbc4bbd337e62c926c995e66a231e5e2b0a722a8`**. That admin can replace the token's code. I did not read who owns it. Claus itself says "token, hook and liquidity contracts support upgrades around the same token address" and "CLAUS DEV retaining owner recovery". The power that matters for a bond is the upgrade key, not only `owner()`. | [S] S8, S3; last sentence [I] | | 5 | Mint authority and transfer pause | `paused()` and `cap()` both revert on the token, so there is no standard pause getter or cap getter. Whether a mint or block function exists I did **not** establish (implementation not read; explorer API was behind a Cloudflare challenge). Because the token is upgradeable, any such power can be added later by the proxy admin. | [S] S8 for the reverts; rest [U]/[I] | | 6 | Supply | `totalSupply()` = **983,872,663 CLAUS** (rounded down) at Robinhood Chain block 112,496,406. The retired Ethereum token `0x1b54E762aa34CF6E28E9C082F2848e28E45DA6b8` still reports 709,042,947 and is still transferable. | [S] S8, S9, S4 | | 7 | "Claus Jobs escrow runs on Ethereum with the old token" | S1 says "Jobs escrow in ETH or CLAUS on Ethereum". Claus calls the Ethereum token "retired". The Jobs page needs JavaScript and I could not read the escrow terms. Not confirmed, not refuted. | [S] S1, S2; status [U] | | 8 | Disputes go to five IMD members where four must agree | Stated on S1. IMD's own skill file describes a configurable oracle: `panelSize` and `quorum` are set per question, the example uses 5 of 5, and "a panel that disagrees ends with no answer, and the payment is still spent". 4-of-5 for Claus deals is not in the IMD documents I read. | [S] S1, S6; mismatch [I] | | 9 | Three reviews cut houses-as-seats and burn-for-a-permanent-seat | S1 says the swarm cut "turning all 300 houses into seats and using a flat CLAUS price". The burn-for-seat cut is not on the page I read; the three job links are given there but I did not open them. | [S] S1; second cut [U] | | 10 | IMD has 2,000 NFT seats | Confirmed: `identity.md` at `0x0000ec93127baa929e58e97dd0095a2bfb38ec1d` on Ethereum, `totalSupply()` = 2000, "free, one per wallet" at mint in May 2026. | [S] S5, S9 | | 11 | "No burn" vs the design's "whatever is left burns" | The public v0.3 page says "no burn in the pilot" and "reviewers are paid regardless of verdict". The design under test adds a burn and a slash. This is a change from the public version, not a continuation. | [S] S1; [I] | | 12 | "No treasury cut" | A customer who buys CLAUS in the official pool to pay a dollar quote pays the 2% pool fee, half of which goes to CLAUS DEV. An arm that sells its pay does the same. So each check carries a treasury cut of up to about 1% on the way in and 1% on the way out unless payment bypasses the pool. The claim holds only inside the fee waterfall. | [S] S3 for the fee; conclusion [I] | --- ## 2. Recommended design ### A. Is a returnable bond plus a trial the right seat? **Verdict [P]: yes for the seat, no for "no cap on arms".** A bond is the only one of the five options that puts the operator's own loss behind a wrong fact. But open entry with no emissions fails when work per arm falls below the cost of staying ready. | Option | What it gives | What goes wrong | Mark | |---|---|---|---| | NFT seat (IMD) | Bounded set, easy scheduling, tradable | Seats concentrate: 2,000 seats minted one per wallet now sit in 789 addresses, the top 10 hold 19.25%, the top 100 hold 57.85% (section 4). Nothing is lost for a wrong answer except future work. | [S] S9 measurement; last sentence [I] | | House based seats (300 NFTs) | Rewards early buyers | Already cut by the swarm (S1). 300 houses in 148 wallets; a house is property, not capacity. | [S] S1, S4 | | Open permit, no bond | Lowest barrier | Fresh wallets are free, so "one controller, one voice" is unenforceable and a wrong verdict costs nothing. | [I] | | Delegated staking | More capital behind each operator | Capital flows to a few operators and delegators are passive. The Graph requires 100,000 GRT self-stake, caps delegation at a multiple of it and uses a 28-day thawing period; Livepeer limits the active set to 100 orchestrators, and rewards there have come mostly from inflation rather than fees. It also pays a yield on idle capital, which rule 3 forbids. | [R] https://thegraph.com/docs/en/indexing/overview/ , https://docs.livepeer.org ; last sentence [I] | | Bond plus trial | Loss for a provable error; trial filters skill | See failures below | [P] | What happened elsewhere, and what I take from it: - **Verifier's dilemma.** If checking costs effort and errors are rare, rational checkers stop checking. Truebit answered with "forced errors", deliberately planted wrong results with a jackpot. The design's planted defects are the same remedy and must be frequent enough to matter. [R] Luu et al. 2015, https://eprint.iacr.org/2015/702 ; Truebit paper, https://people.cs.uchicago.edu/~teutsch/papers/truebit.pdf - **Token-vote arbitration is capturable by large holders.** In March 2025 a Polymarket market resolved through UMA's token vote was reported as swung by one large UMA holder. A slashing decision must not be settled by CLAUS-weighted voting. [R] UMA docs https://docs.uma.xyz ; the incident was widely reported, I did not re-fetch a report. - **Low fees leave jurors and reporters absent.** Augur v1 suffered from thin reporting and "invalid" market griefing; Kleros relies on juror fees paid by the parties and appeal deposits that double each round. Someone must pay the panel on every dispute. [R] https://docs.kleros.io ; Augur whitepaper https://arxiv.org/abs/1501.01042 - **Large bonds close the door.** Arbitrum BoLD set its assertion bond at 3,600 ETH to stop delay attacks, which in practice means few parties post it. Small bonds with capped liability are the right end for a $40 product. [R] https://docs.arbitrum.io/how-arbitrum-works/bold/gentle-introduction - **Unbonding delays in live networks:** Chainlink staking v0.2 uses a 28-day cooldown; EigenLayer uses a 14-day withdrawal escrow. [R] https://blog.chain.link/chainlink-staking-v0-2-overview/ , https://docs.eigenlayer.xyz - **ERC-8004** already describes identity, reputation and validation registries for agents, with stake-secured re-execution named as one validation model. IMD registers seats under it (S1). Arms should register there too rather than invent a new registry. [R] https://eips.ethereum.org/EIPS/eip-8004 ; [S] S1 for IMD's use; last sentence [P] ### B to D. Parameters, each with a number and a reason All rows are **[P]** unless marked. | Parameter | Number | Reason | |---|---|---| | Bond unit | Dollar target, posted in CLAUS, converted at a 7-day average of the official pool, with a **collar of 25,000 to 100,000 CLAUS** | A pure CLAUS amount makes the deterrent drift with the token; a pure dollar amount forces constant top-ups. The collar bounds both. I did not read a CLAUS price, so I cannot say where the dollar target sits inside the collar today. | | Bond, class (a) control check and (c) watch | **$600** | 15 times the $40 check price; one bond backs 10 open checks at a $60 slice each (below). | | Bond, class (b) upgrade review | **$3,000**, max 3 open reviews | Reviews are priced about 6 times higher and judgement is harder; fewer, better-backed arms. | | At-risk slice per open check | **$60** (1.5 times the check price) | Covers a full refund ($40) plus a challenger reward, so a proven error is always fully funded from one slice. | | Open work per bond | **10 checks** | $600 / $60. No bond is ever promised twice. | | Challenge window | **7 days** after publication | Class (a) facts are readable at a pinned block in minutes; 7 days is long enough for a weekend and short enough to recycle a bond four times a month. | | Tail after verdict | Slice stays locked 7 days; if challenged, until the ruling | Same as the window. | | Unbonding delay | **14 days**, no new work once requested | Window (7) plus panel time (7). Matches EigenLayer's 14 days [R]; shorter than the 28 days at The Graph and Chainlink [R] because liability here is capped and short-lived. | | Revaluation | Weekly. Below 80% of target: open-work limit shrinks pro rata. No forced liquidation. | A falling token should reduce an arm's capacity, not eject it. | | Offsets for volatility, no emissions | (1) pay is quoted in dollars and converted at settlement; (2) bond is kept near one month of expected earnings for a working arm; (3) the collar; (4) a cap on active arms so each bond earns | An operator bonds a volatile asset only if work repays it quickly. The lever that exists without emissions is utilisation. | | Active arms | **Capped at 30 in the pilot** with a public waiting list; add 10 when average arms complete more than 20 paid checks a month | Matches S1's "5–10 operators (30 max)". Demand evidence in section 4 supports single digits of arms today. | | Trial | 10 planted cases, at least 9 right, zero wrong "all clear" on a planted defect; repeated every 90 days | Filters skill, cheap to run. Failing the trial is never slashed. | | Planted defect rate on live work | **1 job in 10** | With a $60 slice, expected loss for a lazy answer is about $6 per job, above the few dollars of inference it saves. Lower rates do not deter [I]. | | "Honest unknown" | Never slashed, but unpaid when the other arm and a tie-break arm both read the fact; more than 20% unknowns on planted cases suspends the arm | Otherwise "unknown" is the profit-maximising answer to everything. | | Slash split, customer job | Customer: full refund first ($40). Challenger: **30%** of the slice ($18). Burn: the rest ($2 at these numbers). | Customer whole first, as designed. Challenger share under 50% so an arm that challenges itself through a second wallet still loses at least 70%. The burn is deliberately small: the bond is there to repay, not to shrink supply. | | Slash, planted case | Whole slice to the reserve, no burn | The reserve pays for planted cases; let failures fund it. | | Strikes | 3 proven errors in 90 days: whole bond slashed, arm removed | Repeat errors signal an operator, not an accident. | | Challenge deposit | **$20** in CLAUS, lost if the challenge fails (panel cost first, remainder to the challenged arm) | Half the check price: enough to stop spam, small enough for an outsider. | | Who pays the panel | The loser: failed challenger from the deposit, slashed arm from the slice; the reserve only fronts it | IMD's oracle spends the fee even when the panel disagrees (S6), so someone must carry that cost every time. | | Panel rule | Ask for a **5 of 5** deterministic chain read at a pinned block; no-answer means no slash and deposit returned less the panel fee | Class (a) facts are objective; if five agents cannot agree, the fact was not safe to slash. | | When IMD built the contract | Step 1: three non-IMD arms drawn at random, 3 of 3, reading at the pinned block. Step 2 if contested: an outside juror court named in advance (for example Kleros [R]); the founders must pick it. | Removes IMD from judging its own work. I do not recommend a CLAUS or IMD token vote (see UMA above). | | Check price, class (a) | **$40** | Waterfall below. | | Upgrade review price, class (b) | **$250** (two arms at $90, reserve $45, coordination $25) | Scaled from (a) by effort; advisory unless a timelock exists, as designed. | | Watch price, class (c) | **$10 per contract per month** | Mostly automated diffing; alerts are flags, never slashable verdicts. | | Coordination | Capped at **10% of price**, published monthly with receipts | "At cost" with no cap and no audit is a treasury cut under another name. | | Reserve | **20% of price** until it holds the larger of $5,000 or three months of planted-case and dispute spending | Planted cases alone cost $2.40 per paid check (below). | | Burn start | **Not in the pilot.** Start only after 500 paid checks and a full reserve; until then the remainder goes to the reserve | Agrees with S1 ("no burn in the pilot"). Burning before refunds are funded moves money from customers at risk to holders. | **Waterfall for one $40 control check [P]:** | Line | Amount | Note | |---|---|---| | Arm 1 | $12 | Inference on the arm's own plan plus operator margin [I] | | Arm 2 | $12 | Same; working alone, sealed | | Reserve | $8 | Of which $2.40 funds the 1-in-10 planted job (2 arms at $12, spread over 10 paid jobs); the rest covers tie-break arms on mismatches, refunds and fronted panel fees | | Coordination | $4 | Assigning, sealing, publishing | | Remainder | $4 | To reserve during the pilot; burned afterwards | Is "arms first, reserve, coordination at cost, remainder burned" sustainable? **[I]** Only if the price is set from this cost stack and the arm count is capped. At $24 of arm pay per check, the burn can never be more than about 10% of revenue; it is a by-product, not the business. **Statements safe to make slashable [P]:** only values that anyone can re-read from chain state at a pinned block: `owner()` and role holders; proxy admin and implementation addresses; whether an address has code; timelock address and its minimum delay; whether mint, pause, blocklist or fee-setter selectors exist in verified source and who may call them; fee bounds readable from storage; where LP tokens or positions sit and any lock expiry. **Never slashable [P]:** "safe", "no backdoor", intent, who is behind a wallet or a multisig's signers, behaviour of unverified bytecode, anything off chain, any upgrade review opinion, any watch alert. **Seals [P]:** commit a hash of the form, reveal after both commits. If an arm fails to reveal within 24 hours it forfeits pay for that job and a third arm is drawn. ### E. What Claus must lock before any bond is credible [P] A bond in an upgradeable token is only as good as the upgrade key (corrected fact 4). 1. The proxy admin `0xcbc4…22a8` and the owner chain `0x979D…Dc46` → `0xBb14…F1C5` behind an on-chain timelock of at least **14 days** (the unbonding delay), with every queued change public. 2. A published statement, checkable in code, of whether mint, pause, blocklist or transfer-fee functions exist, and who can call them. This is exactly a class (a) control check; Claus should be its own first customer, checked by IMD, as S1 already proposes. 3. The bond escrow contract immutable, with no owner able to move bonds, and exempt from any transfer restriction the token could gain. 4. An exit rule: if a token upgrade is queued, arms may unbond with no delay beyond open slices. 5. Hook fee setters bounded so the 2% cannot be raised on arms entering or leaving. ### F. Demand - First buyers **[I]**: (1) people launching through IMD. IMD's workflow already requires that a request "ask for an independent security review before deployment, or the check blocks it" ([S] S6), so a control check has a natural slot. (2) Buyers of tokens launched on IMD or Claus who want a neutral reading of who controls the contract. (3) Claus itself, for its own authority map. (4) Projects with a timelock that want an upgrade review. - Price points **[P]**: $40 control check, $15 re-check of an already checked contract, $10 a month watch, $250 upgrade review. ### G. The 300 NFTs gain nothing new - **Strongest objection [I]:** "We paid first and carried the risk. The houses were sold with earnings attached (staking rewards, the 50,000 CLAUS backing). Now the first real revenue line is built so that we are shut out, and it is presented as fairness." - **Best honest answer [P]:** Nothing a house holds is taken: the backing was paid out in full (15 million CLAUS, [S] S4), earlier earned rewards stay payable ([S] S3), and the house stays a house. The new work pays for work, not for ownership, and it is open to house holders on the same terms as anyone. A house holder who wants to earn can bond and pass the trial; the 50,000 CLAUS already paid per house falls inside the proposed bond collar. - **Courtesy that is not income [P]:** a house may name one delegate (already in S1), a house mark on that arm's public profile, and a free first trial attempt. **Not** job priority, **not** a place ahead in the waiting list, **not** a fee share: each of those is income by another route. ### H. One controller, one voice It **cannot be proven** on chain: fresh wallets are free **[I]**. So the design should make hiding unprofitable rather than claim to detect it **[P]**: 1. **Random pairing** from the active pool after orders are placed. With 30 arms, a controller hiding 3 arms gets both seats on a job with probability (3/30)×(2/29) = 0.69%; hiding 10 gives 10.3%. Each hidden arm costs a full bond. 2. **Cheapest workable test:** a signed declaration of common control, plus a public funding-trace rule: arms whose bond or gas traces to the same funding wallet within two transfers count as one controller. It is evadable through an exchange, so it is a screen, not proof. 3. **A bounty:** anyone who proves undeclared common control (same funder, same payout address, same signing key) receives 30% of both bonds; the rest burns. 4. **Planted cases** catch the thing that matters: two colluding arms that copy a lazy answer both lose a slice. --- ## 3. What I would cut or change 1. **Cut "no cap on arms".** Cap at 30 until work per arm justifies more. Evidence: section 4 demand; [R] networks where fees did not cover work. 2. **Cut burning from the pilot.** Reserve first. The burn at these prices is about $4 a check. 3. **Cut "the rest burns" as the main destination of a slash.** Customer and challenger take nearly all of it; planted-case slashes refill the reserve. 4. **Cut "coordination at cost" without a cap.** Cap at 10%, publish receipts. 5. **Change "honest unknown is never punished"** to "never slashed, but not paid when the fact was readable, and suspended above 20% on planted cases". 6. **Cut slashing for classes (b) and (c)** at launch. Only class (a) form fields are slashable. 7. **Change "IMD oracle panel" from 4 of 5 to 5 of 5 chain reads** for slashing, with no-answer meaning no slash. 8. **Change "IMD seats skip the trial".** A seat holder is not an operator of record: 789 addresses hold the 2,000 seats and one holds 77. Let them skip the queue for the trial, not the trial. 9. **Do not launch any bond before the upgrade key is behind a timelock** (section 2E). 10. **Say plainly that "no treasury cut" excludes the 2% pool fee**, or settle outside the pool. 11. **Settle on one chain.** Fees in CLAUS on Robinhood Chain need a new escrow there; the existing Jobs escrow is described as on Ethereum (fact 7). --- ## 4. Measurements and their limits ### E. CLAUS bonded at 10, 100 and 1,000 arms Supply source: `totalSupply()` on `0x163015276f3244bC2CcB78b4D69cc8A4069EfFAC`, Robinhood Chain block 112,496,406 = 983,872,663 CLAUS **[S] S8**. Arithmetic **[I]**. One class (a) bond per arm. | Bond per arm (collar) | 10 arms | 100 arms | 1,000 arms | |---|---|---|---| | 25,000 CLAUS (floor) | 250,000 = 0.025% | 2,500,000 = 0.25% | 25,000,000 = 2.54% | | 50,000 CLAUS (middle) | 500,000 = 0.051% | 5,000,000 = 0.51% | 50,000,000 = 5.08% | | 100,000 CLAUS (ceiling) | 1,000,000 = 0.10% | 10,000,000 = 1.02% | 100,000,000 = 10.16% | For scale: the 15 million CLAUS paid to NFT wallets is 1.52% of this supply **[I]** from S4 and S8. Limits: I did not read a CLAUS price, so I cannot place the $600 target inside the collar. Bonds are returnable, so these are locked amounts, not removed supply. With my recommended cap of 30 arms the middle case is 1,500,000 CLAUS, 0.15% of supply. At that scale the bond has no material effect on circulating supply, and it should not be presented as if it had. ### F. Volume - **IMD launches [S] S7:** the IMD explorer token list showed "1 – 20 of 712" with the 20 newest entries spanning 17 hours to 8 days old. That is about **2.5 new tokens a day** over the last 8 days **[I]**. At least 4 of the 20 were labelled tests ("DO NOT BUY", "Test") and several were duplicates of one project, so distinct real launches are nearer 1 to 2 a day **[I]**. The list mixes Ethereum, Robinhood Chain and Sepolia. - **IMD public repositories [S] S10:** the `Identity-md` GitHub organisation has 8 public repositories (worker, desktop, templates, research), created between Aug and Oct 2026. I did not find a public per-launch repository feed in this run, so the count above comes from the explorer, not from launch repositories **[U]**. - **New tokens a day on Robinhood Chain: not measured [U].** The official explorer (robinhoodchain.blockscout.com, per S11) answered API requests with a Cloudflare browser challenge, and I did not go around it. The public RPC cannot list token creations cheaply. - **What that volume supports [I]:** if 30% of roughly 2 launches a day bought a $40 check, that is about 18 checks a month, about $430 a month of arm pay in total. That feeds a handful of arms, not 100. Demand, not bond design, is the binding limit. ### H. Concentration **IMD seats [S] S9, measured.** Method: `ownerOf(id)` for ids 0 to 2000 on `0x0000ec93127baa929e58e97dd0095a2bfb38ec1d`, Ethereum block 26,398,605; id 0 reverts, ids 1 to 2000 all resolved. | Measure | Value | |---|---| | Distinct holder addresses | 789 | | Addresses holding exactly 1 seat | 567 | | Addresses holding 10 or more | 38 | | Largest holder | 77 seats, 3.85% (`0xb7a832db45f240f876432bf8b222bff30b4be8ff`) | | Next four | 54 (`0xb64199cd690aa4e78f1a6625660347ce63031d72`), 47 (`0x80271ce20184e38f4afe90d4ca134304d197aca2`), 34 (`0xe3de112ae639b45b7d22c4e42742020db679d80c`), 30 (`0xd4bb7351a9b0d1b236e7523da7a972db4a100f08`) | | Top 5 / 10 / 20 / 50 / 100 | 12.1% / 19.25% / 29.7% / 45.8% / 57.85% | | Herfindahl index | 0.0069 | Limits: addresses are not controllers; I did not check whether the top addresses are contracts (a vault or marketplace escrow would overstate concentration), and one person with many wallets would understate it. **Claus NFTs [S] S4, not independently measured.** Claus states the 300 NFTs' payment went to 148 wallets with staked NFTs attributed to stakers: a mean of 2.03 houses per wallet **[I]**. I did not locate the NFT contract address in time to read owners, so top-holder shares are **[U]**. **CLAUS holdings on Robinhood Chain: not measured [U].** Method attempted: sum `Transfer` logs for the token from the public RPC. The RPC limits a log query to 10,000,000 blocks and my requests failed before completing a scan of 112 million blocks; the explorer's holder API was behind a browser challenge. What is known from Claus's own page **[S] S4**: 3,805 direct allocations and 141 gate migrations were delivered; the final liquidity deposit added 11,004,285.57 CLAUS to the pool; CLAUS DEV kept 3,007,091.08 CLAUS unused. To finish this measurement: page the Transfer logs from the token's deployment block in 10M-block steps, net the balances, and report top 1/10/100 shares excluding the pool manager and burn address, with the block number. --- ## 5. Open questions only the founders can answer **Claus** 1. Who owns the proxy admin `0xcbc4bbd337e62c926c995e66a231e5e2b0a722a8`, and will it and the owner chain go behind a timelock of at least 14 days? 2. Does the current token code contain mint, pause, blocklist or fee-on-transfer functions, and who can call them? 3. Will a check be paid outside the 2% pool, or does Claus accept that the pool fee makes "no treasury cut" untrue at the edges? 4. Will Claus deploy a new escrow on Robinhood Chain, or keep Jobs escrow on Ethereum? In which token does Jobs escrow settle today? 5. Will Claus publish its coordination cost with receipts and accept a 10% cap? 6. Who funds the reserve before the first 500 checks? 7. What exactly was promised to NFT buyers about future income, in writing? **IMD** 8. Will the oracle panel accept slashing disputes at 5 of 5, and at what fee? Is the 4-of-5 rule for Claus deals written anywhere public? 9. Who arbitrates when IMD built the contract, and will IMD accept an outside court named in advance? 10. Will IMD's launch workflow accept a Claus control check as the "independent security review" it already requires? 11. Can a seat holder with many seats be treated as one controller for arm purposes? **Both** 12. What is the maximum loss a customer can claim for a wrong check: the fee, or more? 13. What is the number of real, non-test launches a day on each side, from their own logs? --- ## 6. Five sentences for a CLAUS holder who is tired of the NFTs earning 1. An arm is not an NFT and not for sale: anyone can grow one by locking CLAUS as a bond, passing a test, and doing the checking work on their own machine. 2. An arm earns only when it works, and a check counts only when two arms that cannot see each other's answers report the same facts. 3. If an arm states a checkable fact that is proven wrong, its bond first repays the customer, then pays whoever proved it, and what remains is burned. 4. The 300 houses keep everything they already have and get no share of this: no fee cut, no reserved seats, no new payout. 5. No fee goes to a treasury or to the houses; arms are paid, a reserve for refunds is filled, and once that reserve is full whatever is left over is burned, which touches every holder's CLAUS in the same way. --- ## Limits of this view - Sources marked [R] were not re-fetched; their numbers should be checked before anyone relies on them. - Three measurements asked for are missing: CLAUS holder concentration, Claus NFT holder distribution beyond Claus's own count, and new tokens a day on Robinhood Chain. Reasons are given in section 4. - I did not read a CLAUS price, by choice and by brief; every dollar figure is a proposal, not a market reading. - I did not open the three IMD review jobs linked from S1 or the Claus Jobs page (JavaScript only). - Chain reads are single-RPC reads at the stated blocks and were not cross-checked against a second provider.