# Claus arms: a CLAUS work bond instead of NFT seats. Researcher 5's independent view Date: 2026-11-12. Desk research plus read-only chain reads. No login, and no write or private endpoint was used. Labels: **[R]** read from a cited source or chain read · **[I]** inferred by me · **[P]** my own proposal · **[U]** unverified or uncertain. Chain reads were made against the public chains (Robinhood Chain 4663, public RPC `https://rpc.mainnet.chain.robinhood.com`; Ethereum mainnet) at Robinhood block **112,496,734** (timestamp 1794483406) and Ethereum block **26,398,617**. Anyone can repeat them with `eth_call`, `eth_getStorageAt` and `eth_getLogs`. Sources used: - S1 claus.si home: https://claus.si/ - S2 Claus migration guide: https://claus.si/docs/migration - S3 Claus NFT access doc: https://claus.si/docs/nft-inference - S4 earlier proposal page: https://swarmcard.imd.site/ - S5 IMD docs: https://imd.fun/docs - S6 Identity-md GitHub org: https://github.com/Identity-md - S7 Robinhood Chain explorer is Blockscout at robinhoodchain.blockscout.com, per https://docs.bitquery.io/docs/blockchain/robinhood/robinhood-chain-rpc-chain-id/ (the explorer's API sat behind a Cloudflare challenge, so I rebuilt holder data from logs) - S8 third-party NFT aggregator: https://egamers.io/nft-collections/contractclaus/ (low trust) - C1–C9: background on real networks, listed in §7. I cite these from prior knowledge and **did not re-fetch them in this run** **[U]**. --- ## 1. Corrected facts | # | Claim in the brief | What the record shows | Label | |---|---|---|---| | 1 | CLAUS moved to Robinhood Chain, token `0x1630…fFAC` | Confirmed. `symbol()`=CLAUS, `decimals()`=18. The Ethereum source token is `0x1b54E762aa34CF6E28E9C082F2848e28E45DA6b8`, and claus.si calls it "retired". | R (chain, S1, S2) | | 2 | Pool fee 2%, half buyback-and-burn, half treasury | Confirmed: "2% project swap fee: 1% funds buybacks and burns, 1% goes to the CLAUS DEV treasury". Automatic liquidity additions have stopped. | R (S2) | | 3 | Each of 300 NFTs got 50,000 CLAUS, 15M to 148 wallets | Confirmed. The payment came from the existing reserve, not from new minting. Staked NFTs were paid to their stakers at snapshot block 26,357,823. | R (S2) | | 4 | "The token is owned by a contract owned by one plain wallet" | **True but incomplete.** (a) `owner()` = `0x979D5810…bDc46`, a contract whose `owner()` = `0xBb145cA8…cF1C5`, which has no code (an EOA, nonce 240). (b) **The token is also an upgradeable ERC‑1967 proxy.** The implementation slot holds `0xd68727dd…14bff` and the admin slot holds `0xcbc4bbd3…722a8`, whose `owner()` is the same `0x979D…` contract. The one EOA can therefore **replace the token's code**. (c) The implementation exposes `mint(address,uint256)`. A simulated `eth_call` of `mint` from `0x979D…` succeeds. From any other address it reverts with `OwnableUnauthorizedAccount`. **Minting is live and controlled by that one EOA.** (d) I found no `pause()` selector in the current implementation, but an upgrade could add one. (e) The 300-NFT contract on Ethereum (`0x80396c71…a95d9`, name "Claus", totalSupply 300) has `owner()` = the same EOA `0xBb14…`. | R (chain) | | 5 | Claus Jobs escrow runs on Ethereum with the old token | **Conflicting sources.** S4 says Jobs escrow pays "in ETH or CLAUS on Ethereum". S1 says the old Ethereum token is retired. I found no escrow address. Escrow that still settles in a retired token would be a defect in its own right. | R (conflict) / U | | 6 | Disputes go to a panel of five IMD members, four must agree | Stated on S4 only. The IMD docs describe "oracle: a signed answer from a panel of agents" but no dispute process. | R (S4, S5), U | | 7 | (implied) NFTs are "earning" | Claus docs say **NFT-sponsored inference has ended** (`410 Gone`, `nft_sponsorship_retired`), and city buildings no longer need an NFT. The 2% fee split in S2 has no NFT share. One aggregator (S8) claims NFT holders share 0.35% of pool volume in ETH and can redeem 50,000 CLAUS. **I could not confirm this from any Claus page or contract.** The NFTs' clear recent gain was the one-off 15M CLAUS. | R (S2, S3), U (S8) | | 8 | Supply (not in the brief) | `totalSupply()` = **983,872,663.40 CLAUS**. Summing every Transfer log since deployment (block 107,148,941; 14,877 transfers) gives the same figure. Balance at `0x…dEaD` is 0, so burns appear to reduce `totalSupply`. Because `mint` is live, supply is **not capped**. | R (chain) | --- ## 2. Answers A–I ### A. Is a returnable bond plus a trial the right seat? - **[I]** Yes, with conditions. It is the only option of the five that puts capital at risk *per verdict*, and that matters for a product whose output is a claim someone relies on. - **NFT seats (IMD style):** a fixed set of 2,000 seats spread over 789 owners (§4H). Nothing is at risk when a seat is wrong. - **House seats:** already cut by three IMD reviews (brief). The houses are also concentrated: one contract holds 185 of the 300. - **Open permits with no bond:** Sybil-free only if identity is strong, and nothing pays the customer back when a check is wrong. - **Delegated staking:** brings capital in but tends to concentrate on a few operators, with apathetic delegators. On The Graph a small set of indexers draws most delegation (C5). On Lido/Ethereum, stake pools toward a few operators (C6). - **Bonded challenge games in practice [U, from C1–C4]:** - UMA's optimistic oracle works for cheap, frequent assertions. Its escalation vote can be swayed by large token holders: the 2025 Polymarket "Ukraine minerals deal" dispute was widely reported as whale-decided (C1). - Truebit's "verifier's dilemma" shows that if checkers are paid only when they find faults, nobody checks. Planted defects ("forced errors") were Truebit's answer, and this design reuses it (C2). - Kleros juror staking works but sees low participation in low-value courts (C3). - Chainlink staking v0.2 capped pools filled within hours, mostly from large wallets (C4). - Optimism and Arbitrum fault proofs (C7, C8) show that permissionless challenges need **bonds sized to the cost of a challenge and a hard time window**. Arbitrum's BoLD had to raise its assertion bond to resist delay attacks. - **[P]** Keep the bond plus trial. Add a **per-job exposure limit** (B) so capital concentration buys more parallel jobs, not more votes. Add **paid planted defects** so checking pays even when nothing is wrong. Do not add delegation in v1. ### B. Bond size - **[P] Unit:** post the bond in CLAUS but *size the risk in dollars*. The base bond is **250,000 CLAUS** per arm (5× one NFT's migration payment, 0.025% of supply). - Why CLAUS units: a fixed CLAUS figure is predictable for operators and needs no price feed to join. - Each job carries a **dollar exposure** (§D). An arm may hold open jobs only while the bond's value at a **7‑day TWAP** of the official pool covers the sum of its open exposures. - **[P] Per job class.** Exposure is customer refund plus challenger reward: - control check: exposure **$500** - upgrade review: **$2,000** - watch: **$100 per month** per watched contract - **[P] How long liability lasts after a verdict:** control check **30 days**. Upgrade review: until the upgrade executes plus **14 days**, capped at 90. Watch: liability for each alert lasts 30 days. - **[P] Unbonding delay: 30 days**, and never shorter than the arm's longest open liability window. It also has to be shorter than any token-governance timelock (E), so arms can exit before a hostile change takes effect. - **[I] Volatility:** a falling token shrinks capacity, so the arm takes fewer jobs. **It is not liquidated.** That removes the forced-sale spiral without any emissions. Operators still carry token risk, and some will not bond for that reason. - **[P] Offsets without emissions:** - top-up grace of 7 days - fees quoted in dollars, so income does not fall with the token - optionally, part of the bond in **ETH** (up to 50%), slashed by the same rules. This lowers token exposure, though it weakens "upside through the token". Founders must decide (§5). ### C. The challenge game - **[P] Safe to slash:** facts that can be read deterministically from chain state at a stated block. Examples: - `owner()` and admin-slot values - whether `mint` or `pause` selectors exist and who can call them (by simulation) - proxy/implementation addresses - timelock delay value - who holds LP tokens or positions - fee-setter roles - **Never slashable:** severity opinions, "no vulnerability", economic judgements, and "unknown" answers. "Unknown" is never punished (brief, point 4). - **[P] Split of a slashed exposure:** 1. **customer first**, up to 100% of the fee plus a fixed make-good of 2× the fee 2. **challenger 30%** of the rest, at least the challenger's own bond back plus 2× the challenge cost 3. remaining **burn** - Slash only the job's exposure, not the whole bond, unless the arm is found to have colluded. - **[P] Anti-griefing:** - The challenger posts a bond equal to **20% of the exposure** and loses it if wrong. - Challenges must name the exact field, block and evidence (a call trace or storage read). - One open challenge per field. - **[P] Anti-collusion** (an arm challenging its twin to collect the reward): - pay the challenger from the burn share only after a cooldown - deny rewards when the challenger and the slashed arm share a funder within 2 hops (H) - the matching second arm is also liable when both landed on the same wrong fact. Both pay, which deters copying answers between paired arms. - **[P] Who pays the panel:** the loser of the challenge, out of the bond, at a fixed **$150 per panel question**. The reserve covers it when the chain itself settles the fact and no loser exists. - **[P] When IMD built the contract:** most slashable facts are chain-readable, so first use **automatic resolution**: a published script re-reads the state at the named block. If a question still needs judgement, use a non-IMD panel. Options are **UMA's optimistic oracle** or a **Kleros court** (both public and paid per case), or three arms with no IMD seat chosen at random from the bonded set. Founders choose (§5). ### D. Fee flow and price - **[I]** "Arms first, reserve, coordination at cost, burn the rest" can be sustained only if the price covers two independent runs plus planted-defect overhead. Underpricing is the failure that shows up across networks (C2, C3). - **[P] Cost build for one control check (the main product):** | Line | $ | Reason | |---|---|---| | Arm A pay | 35 | ~20–40 min of an operator's agent run on own AI plan, plus review | | Arm B pay | 35 | independent second run | | Planted-defect overhead | 10 | ~1 planted job per 8 real ones, paid at full rate | | Reserve (refunds, disputes) | 15 | reserve target below | | Coordination/publishing | 10 | assignment, sealing, onchain record | | Remainder to burn | 4 | | | **Price** | **$109 → list at $99–$129** | | - Upgrade review: **$400–$600**. - Watch: **$20 per contract per month**. - **[P] When to start burning:** only once the reserve holds **the greater of $25,000 or 3 months of refunds and panel costs**. Until then the remainder goes to the reserve. Publish both balances monthly. ### E. Token effect (no price talk) - **[R]** Supply 983,872,663.40 CLAUS at Robinhood block 112,496,734, from `totalSupply()`. - **[P → I]** At 250,000 CLAUS per arm: | Arms | CLAUS bonded | Share of supply | |---|---|---| | 10 | 2,500,000 | 0.25% | | 100 | 25,000,000 | 2.54% | | 1,000 | 250,000,000 | 25.4% | - **[R → I]** 766 addresses hold ≥50,000 CLAUS and 172 hold ≥1M (§4H). About 172 addresses could each fund several bonds today. 1,000 arms would mean either buying or heavy pooling. I read 1,000 as unrealistic in year one. - **[I] What hurts the bond's credibility:** - one EOA can mint without limit, which dilutes every bond - it can upgrade the token to add a pause, freeze or blacklist, trapping bonds and challenger rewards - it controls the NFT contract - **[P] What Claus must lock before any bond is taken:** 1. Move ownership of `0x979D…` (and with it the ProxyAdmin) to a multisig behind a **timelock of at least 35 days**, longer than unbonding. 2. Either renounce `mint` or cap it in code with a published schedule. 3. Hold bonds in a **non-upgradeable** bond vault with immutable slashing rules. 4. Publish the vault and timelock addresses. ### F. Demand - **[R/I] Robinhood Chain new pools a day.** I counted pool-creation events across the whole chain via `eth_getLogs` (block time 0.105 s, about 822,700 blocks a day): - Uniswap‑v4‑style `Initialize`: **2,696 in the last day, 3,423/day over 7 days**, almost all from PoolManager `0x8366a39c…40951` - v2 `PairCreated`: **279 in 1 day (773/day over 7)** - v3 `PoolCreated`: **170 in 1 day (162/day over 7)** - **Limit:** pools are not unique tokens. Many are launchpad deployments or duplicates, so treat this as an upper bound on new tradeable tokens. - **[U] IMD launches a day:** not measured. The Identity-md org (S6) has 8 public repositories (worker, desktop, box, research…). None is a per-launch registry I could count. IMD launches would have to be counted from the IMD intake contract's events (§5). - **[I] Likely first buyers:** - launchpads and token deployers on Robinhood Chain who want a "control verified" badge - IMD itself, for its own launches (but see C on independence) - traders before buying - **[P] Price points:** - control check **$99–$129** paid by the project - optional **$9** "re-read" of a published check at a later block - watch **$20/month** - upgrade review **$400–$600** ### G. The 300 NFTs gain nothing new - **Strongest objection [I]:** "Claus's earliest backers are being written out of its most valuable next product, and the same single wallet that controls our NFT contract decides." - **Best honest answer [P]:** - Holders received 15M CLAUS in total a month ago (S2). - Under this design that CLAUS gives them exactly the same rights as anyone's CLAUS. They can bond it and work. - Upside flows to all holders through the burn, not to a class of holders. - **Courtesy that creates no income [P]:** - priority in the **first trial window** (they still pass the same trial and post the same bond) - a cosmetic "founding house" mark on an arm's public profile - Neither affects pay, votes or job assignment. ### H. Concentration **CLAUS on Robinhood Chain** **[R]**, method: replayed all 14,877 Transfer logs of `0x1630…fFAC` from deployment block 107,148,941 to 112,496,734. Positive balances: 3,982 addresses. | Measure | Value | |---|---| | Top 1 | 9.85% | | Top 5 | 21.6% | | Top 10 | 29.9% | | Top 20 | 40.6% | | Top 50 | 59.2% | | Top 100 | 74.0% | | Addresses to reach >50% | 33 | | Gini | 0.953 | | HHI | 178 | - **Largest holders and what they are:** - #1 `0x8366a39c…40951` (96.9M) is the v4 PoolManager, which emits the chain's `Initialize` events. Its balance is pooled liquidity, not one owner **[R/I]**. - #2 `0x63b9a335…` (40.3M) is a small contract (code length 166 bytes). Its type is unknown **[U]**. - #3–#5 have 48-byte code, consistent with EIP‑7702 delegated EOAs **[I]**. - 14 of the top 25 are contracts, so "holders" overstates the number of independent people **[I]**. **300 Claus NFTs (Ethereum `0x80396c71…a95d9`)** **[R]**, method: last Transfer per tokenId from deployment block 26,129,224 to 26,398,617. - 69 owners. - `0x4e591642…08cad4` (a contract, probably the staking vault **[I]**) holds **185 (61.7%)**. - Top 5 hold 73.0% and top 10 hold 77.3%. 50 owners hold exactly one. - The vault's depositors were not resolved **[U]**. S2's "148 wallets" counts stakers. **2,000 IMD seats (Ethereum `0x0000ec93…8ec1d`)** **[R]**, same method from block 25,090,085. - 789 owners. - Top 1 holds 77 (3.9%), top 5 hold 12.1%, top 10 hold 19.2% and top 20 hold 29.7%. It takes 64 owners to pass 50%. - 567 owners hold exactly one. - The top holder `0xb7a832db…` is an EOA. The docs' "one per wallet" (S5) is clearly not enforced at the token level. **Can "one controller, one voice" be enforced? [I]** Not fully. Fresh wallets are free. **The cheapest workable test [P]:** 1. Never pair two arms whose **bond deposits trace to a common funder within 2 hops** in the 90 days before bonding. This is cheap to compute from Transfer logs, as I did above. 2. Never pair arms that **share a payout address**. 3. Never pair arms whose sealed answers are byte-identical in free-text fields. That is a copying signal, and it triggers a review. 4. Randomly assign the second arm from the whole bonded set, so a controller needs a large share of all arms to pair with itself often. At 10% of arms, a controller pairs with itself about 10% of the time. Points 1–3 cut that further. It is not proof against a patient attacker, which is why both matching arms are liable (C). ### I. See §6. --- ## 3. Recommended design (each open parameter) | Parameter | Value | Reason | Label | |---|---|---|---| | Base bond | 250,000 CLAUS | 0.025% of supply, 5× an NFT payment, and 172 addresses can fund it today | P | | Exposure per control check / upgrade / watch-month | $500 / $2,000 / $100 | covers fee refund, make-good and challenger reward | P | | Capacity rule | open exposure ≤ bond value at 7‑day TWAP | turns volatility into lower capacity, not liquidation | P | | Liability window | 30 d (check), upgrade+14 d ≤90 d (review) | a fact at a block is quickly re-checkable | P | | Unbonding | 30 d, never shorter than the longest open liability window | lets challenges land | P | | Trial | 20 planted forms; pass ≥19 with zero false "confirmed" | false confirmations are the costly error | P | | Challenger bond | 20% of exposure | deters griefing but stays affordable | P | | Slash split | customer (fee + 2× fee), then 30% challenger, rest burned | customer first, as the brief requires | P | | Panel fee | $150, paid by the loser | panels are not free | P | | Planted-defect rate | 1 in 8 jobs | Truebit-style forced errors keep checking worth doing | P | | Reserve target before burn | the greater of $25k or 3 months of costs | no burns while refunds are underfunded | P | | Price | control $99–129, upgrade $400–600, watch $20/mo | §D build | P | | IMD seats | skip the trial, post the full bond, are **never paired with each other on an IMD-built contract** | independence | P | | Pre-launch lock | multisig + ≥35 d timelock on token owner/ProxyAdmin, mint cap or renounce, immutable bond vault | otherwise the bond is not credible | P | ## 4. What I would cut - **[P]** Upgrade reviews (product b) in v1: defer them until control checks have a dispute history. When they ship, keep "binding only where the project has a timelock". - **[P]** Watch (product c) as a slashable product: an alert should be advisory, with slashing only for a missed change to a field that was checked. - **[P]** "Anyone can challenge" without a challenger bond: cut. Challenges need a bond. - **[P]** Slashing the *whole* bond for one wrong fact: cut. Slash the job's exposure, and the whole bond only on proven collusion. - **[P]** Settling in CLAUS while Jobs escrow sits on Ethereum with a retired token: cut. Settle only on Robinhood Chain. - **[P]** The IMD panel as the default arbiter. Most facts should resolve by script, with the panel for leftovers only. - **[P]** Any launch of the bond before the §3 lock. As it stands, one EOA can mint and upgrade the token the bonds sit in. ## 4b. Measurements and their limits (E, F, H) - **E:** supply comes from `totalSupply()` and is cross-checked by the log replay. The figure is not fixed, because `mint` is live. - **F:** - Pool counts are pools, not unique tokens. Signatures are standard v2/v3/v4 events, and other DEX designs are missed. - The day count is one sample. The 7-day mean is the better figure. - IMD launches were not measured. - **H:** - Addresses are not owners. Contracts (the pool, the vault, 7702 wallets) were not looked through. - NFT staking-vault depositors were not resolved. - The Blockscout API was blocked by Cloudflare, so there was no cross-check with explorer labels. - **All:** the counts can be repeated from the stated blocks. External references C1–C9 were not re-fetched here. ## 5. Open questions only the founders can answer 1. Claus: will you put the token owner and ProxyAdmin behind a multisig and a ≥35‑day timelock, and cap or renounce `mint`, before any bond is taken? 2. Claus: what is contract `0x4e591642…` (185 NFTs), and does any NFT fee share (S8's 0.35% claim) exist today? 3. Claus: where is the Jobs escrow contract, and which token does it settle in now? 4. IMD: who sits on the five-member panel, how are its members chosen, and is the panel paid? 5. Both: who arbitrates when IMD built the contract: UMA, Kleros, or random non-IMD arms? 6. Both: may part of a bond be ETH, trading less token exposure for more operators? 7. IMD: can launch counts per day be published from the intake contract? ## 6. Five sentences for a CLAUS holder tired of the NFTs earning 1. Under this design the NFTs get nothing new: no fee share, no seats, no special pay. 2. Anyone can become a checker, called an arm, by bonding their own CLAUS and passing a test with hidden planted mistakes. 3. Arms are paid only for checks that a second arm, working alone, confirms, and an arm that states a checkable fact wrongly pays the customer back from its bond. 4. After arms, a safety reserve and running costs are paid, whatever is left of each fee is burned, so every holder shares the result alike. 5. None of this should start until the single wallet that can mint and upgrade CLAUS is locked behind a public delay, because a bond is only as safe as the token it is held in. ## 7. External references (from prior knowledge; not re-fetched in this run) [U] - C1 UMA optimistic oracle docs: https://docs.uma.xyz/ (and press coverage of the 2025 Polymarket Ukraine-minerals dispute) - C2 Truebit whitepaper, "verifier's dilemma" and forced errors: https://people.cs.uchicago.edu/~teutsch/papers/truebit.pdf - C3 Kleros: https://kleros.io/whitepaper.pdf - C4 Chainlink Staking v0.2: https://blog.chain.link/chainlink-staking-v0-2-overview/ - C5 The Graph delegation: https://thegraph.com/docs/en/resources/roles/delegating/delegating/ - C6 Lido node operators: https://lido.fi/ - C7 Optimism fault proofs: https://docs.optimism.io/stack/fault-proofs/explainer - C8 Arbitrum BoLD: https://docs.arbitrum.io/how-arbitrum-works/bold/gentle-introduction - C9 ERC‑8004: https://eips.ethereum.org/EIPS/eip-8004