# Claus arms: an open work bond, with bounded promises Independent view 4 · Public-source desk research · 12 November 2026 **[P — recommendation]** Keep open admission, a returnable CLAUS bond, independent checks and payment for work. Change the financial promise: a volatile CLAUS bond should not also be the customer's dollar refund fund. Start with a $100-equivalent CLAUS admission bond and $250 of stable settlement collateral **per arm per open control check**, a $65 retail control check, and a seven-day challenge period. No new NFT income. Launch only after the token's administrative powers and the dispute signer are independently mapped and constrained. **[I — central criticism]** Two matching agents establish agreement, not truth or independent ownership. A bond establishes a recoverable penalty, not a guarantee against the loss of a customer's assets. The product should sell a reproducible, block-specific description of authority, with a service-fee refund, rather than a safety certificate. ## Evidence conventions and scope **[R]** means read from the linked public source, including source code or public chain-index responses; it describes what that source reports. **[I]** means my inference or calculation from those observations. **[P]** means my proposed rule, budget or experiment, not an existing Claus or IMD commitment. **[U]** identifies an unverified claim or unavailable measurement. These labels apply to the whole paragraph or to every row of a table explicitly labelled with one. Dollar amounts are accounting and service-cost parameters, not investment claims. **[R — method]** Research used public websites, documentation, explorer pages and anonymous read APIs, public GitHub repository metadata, ERC-8004 and operators' published accounts of their own networks. No account login, private endpoint, transaction submission or private repository was used. Public RPC requests attempted only chain reads. No other researcher's submission was consulted. GitHub listing metadata was counted without opening this assignment's repository or other views. **[U — limits]** Some Claus guides, explorer APIs and public RPCs returned HTTP 403; Routescan did not support Robinhood chain 4663. Ethereum NFT measurements below are pinned, reconstructed measurements. The accessible CLAUS holder page is an internally inconsistent, unpinned explorer snapshot: it is evidence to reconcile, not a certified current distribution. The earlier proposal website was inaccessible. This report does not silently replace missing chain observations with the supplied brief. ## 1. Corrected facts and qualifications | Status | Observation and correction | Evidence | |---|---|---| | R | Official CLAUS is Robinhood chain 4663, `0x163015276f3244bC2CcB78b4D69cc8A4069EfFAC`. Ethereum source token is `0x1b54E762aa34CF6E28E9C082F2848e28E45DA6b8`. Migration now uses Ethereum snapshot block **26,357,823**; new migration burns are closed. | [Current migration guide](https://claus.si/docs/migration) | | R | The guide reports **300 × 50,000 = 15,000,000 CLAUS**, delivered to **148 wallets**, with staked NFTs attributed to stakers. The snapshot JSON separately records 300 NFT backing deliveries and transaction `0xbd404685574a53e58e2ebd5b5887486a3330a7b3d32121b4cc8364e6761c9046`. | [Migration guide](https://claus.si/docs/migration), [snapshot record](https://claus.si/api/migration/snapshot), [payment transaction](https://robinhoodchain.blockscout.com/tx/0xbd404685574a53e58e2ebd5b5887486a3330a7b3d32121b4cc8364e6761c9046) | | R | Current project swap fee is **2%: 1% buyback/burn, 1% operating treasury**. This is separate from LP fees and from the proposed checking-service fees. | [Current funding description](https://claus.si/read/Hooks/Funding) | | R/I | The migration article's old body still describes automatic liquidity funding, an open burn migration and remaining Ethereum project liquidity. Its update and the current guide supersede those statements. Do not use the historical article body as a current operating specification. | [Article and update](https://claus.si/migration-article), [current guide](https://claus.si/docs/migration) | | R/U | The article says a single founder wallet controls upgrades, minting, fees and project liquidity, with no enforced mint cap or waiting period. The build text instead says core upgrade authority is unavailable and the owner can change metadata only. The explorer-served implementation contains `mint(address,uint256) external onlyOwner`. These are materially conflicting descriptions. I have **not** verified the brief's precise current token-owner-contract → plain-wallet ownership chain. | [Article](https://claus.si/migration-article), [build text, lines 84–85](https://claus.si/build.txt), [explorer implementation/source](https://scan.codex.io/robinhood/address/0x163015276f3244bc2ccb78b4d69cc8a4069effac?tab=code) | | R/I | `setMigrationPaused(bool)` appears in that implementation; no general transfer-pause function was found in its served ABI. Pausing migration is not pausing transfers. This does not prove that another implementation, proxy administrator or future upgrade cannot introduce restrictions. | [Explorer-served ABI and code](https://scan.codex.io/robinhood/address/0x163015276f3244bc2ccb78b4d69cc8a4069effac?tab=code) | | R/U | The public build text describes Jobs as a free public task board, with paid ETH/CLAUS agreements as a separate rail. Thus “Jobs escrow” needs that qualification. The specific current Ethereum escrow addresses, old-token pin and five-IMD-member/four-vote dispute configuration in the brief were **not independently verified**: their linked detailed guides were inaccessible. | [Build text, Jobs and paid-agreement sections](https://claus.si/build.txt), [public priorities](https://claus.si/Priorities) | | R | IMD currently describes **2,000 NFT seats**, own-machine/own-plan workers, and launch allocations of **8% to connected seats plus 2% to accepted contributors**. Carrying those token allocations into a checking swarm would conflict with the proposed independence rule. | [IMD docs](https://imd.fun/docs/) | | R/I | ERC-8004 uses an ERC-721 identity registry, while validation incentives and slashing sit outside the registry. An arm can therefore have an ERC-8004 identity NFT without making that NFT its scarce work permit. The standard explicitly acknowledges Sybil attacks; registration does not prove independent control. | [ERC-8004](https://eips.ethereum.org/EIPS/eip-8004) | **[P]** Keep historical NFT claims, the migration payment and future swarm revenue in separate ledgers. Do not infer that the Robinhood payment cancelled every Ethereum redemption or accrued-reward obligation. Reconcile those obligations contract by contract before describing what NFT holders “keep.” ## 2. A — Is a bond plus a trial the right seat? **[P — comparison]** | Admission mechanism | Useful property | Main weakness | Decision | |---|---|---|---| | Returnable work bond + trial | Open entry, withdrawable capital, a funded penalty | Wealth gate, volatile collateral, hidden common owners | Best starting point, with per-job collateral and limited claims | | IMD-style scarce NFT seat | Persistent identity and a simple finite registry | Seat ownership does not establish checking skill; transferable scarcity can dominate work access | Do not copy its admission scarcity or launch allocations | | The 300 Claus houses as seats | Existing community and recognizable identities | Inherits existing holdings, custody arrangements and an arbitrary cap of 300 | Reject | | Open permit, no bond | Lowest entry barrier | Cheap abandonment, repeated identities and no funded redress | Allow training and unsigned research, not final paid attestations | | Delegated staking | Operators need less personal capital | Backers seek passive income, incumbents attract capital, responsibility becomes diffuse | Do not add at launch; reassess only after measured demand | **[R — networks actually used]** The Graph requires an indexer self-stake of 100,000 GRT, pays both query fees and issuance-funded indexing rewards, and exposes indexers to slashing. Its delegation documentation allows capacity up to 16 times self-stake and warns that reward cuts can leave delegators nothing. This is an operating bonded-work network, not evidence that fees alone finance the same arrangement. [Indexer documentation](https://thegraph.com/docs/en/indexing/overview/), [delegation documentation](https://thegraph.com/docs/en/resources/roles/delegating/delegating/). **[R/I — apathy and concentration]** The Graph's retrospective reports that 15.2% of 2025 indexing rewards went to indexers not serving queries or indexing consumer-used data. It also reports a single delegator removing 25.69 million GRT from never-renewed indexers, versus 28.37 million total undelegated from failing indexers in that stage. My inference: capital allocation and incentives can remain detached from useful work; a small number of large allocators can dominate the apparent response. This is not a measurement of Claus operators. [Graph retrospective](https://thegraph.com/blog/reo-two-months-retrospective/). **[R/I — fee pressure]** A historical Graph operator FAQ put closing and reopening an allocation at about $50 each in then-current Ethereum gas conditions, warning that daily recycling could exceed infrastructure costs. Those are historical costs, not today's Robinhood costs. Kleros later moved security registries to Gnosis for lower deposits and costs and offered a 300,000-PNK contribution incentive pool. These examples undermine the assumption that posting a bond automatically attracts sufficiently paid work or monitoring. [Graph FAQ](https://forum.thegraph.com/t/delegators-protocol-actions-faq/89), [Kleros programme](https://blog.kleros.io/incentives-program-for-security-registries/). **[R/I — challenge games]** UMA's OOv3 accepts undisputed assertions after a challenge interval and escalates disputed assertions to the DVM. Its tutorial explicitly warns that its bond-free, two-minute example is unsuitable for mainnet. Kleros Stake Curate reports a 2025 deployment using permanent deposits, challenge stakes and separately funded arbitration. These document working mechanisms, not proof of sufficient independent challengers. [UMA tutorial](https://docs.uma.xyz/developers/optimistic-oracle-v3/quick-start), [Stake Curate description](https://blog.kleros.io/cut-compliance-costs-with-stake-curate-the-high-quality-verification-tool/). **[R/I — actual failure]** Kleros's August 2026 agent-court account describes case 173: one agent went offline, one juror held two remaining votes, the result tied, and every juror was slashed. The same account describes a successful five-agent demonstration but labels it a beta test. My conclusion is to prohibit automatic punishment merely for being outside a majority, and to require working timeout and appeal paths before using agent juries. [Kleros account and linked cases](https://blog.kleros.io/kleros-live-stream-26-august-2026-a-dispute-was-filed-during-the-call-and-five-agents-ruled-it-before-it-ended/). ## 3. B — Recommended admission, collateral and exposure **[P — all parameters below]** The crucial change is a **hybrid bond**, not a larger CLAUS-only bond. CLAUS remains the returnable admission stake; stable collateral finances bounded dollar liabilities. This explicitly departs from the proposed single-asset design. | Parameter | Initial number | Reason | |---|---|---| | CLAUS admission bond | $100 equivalent at entry; token quantity fixed for that admission | Makes resetting an identity costly without requiring the work bond to insure arbitrary customer losses | | Control-check collateral | $250 in a named, redeemable dollar settlement asset per arm per job | Covers a $65 service refund, $25 bounty and $50 first panel, with $110 headroom even if only one arm is wrong | | Open exposure | At most 2 control checks per controller, including checks still challengeable | $500 stable collateral cannot be reused across ten apparently independent jobs | | Upgrade class | $2,000 stable collateral per arm; 1 upgrade at a time | Larger evidence and escalation cost; still only a bounded service remedy, not project-asset insurance | | Watch class | $250 per monitored contract per arm; at most 2 contracts per controller initially | A continuing obligation must not leave infinite future alerts secured by one old bond | | Initial trial | 20 hidden fixtures; all 8 critical authority defects found, at least 18/20 overall; correct “unknown” counts as correct when appropriate | Tests reproducibility and abstention, not confidence or agreement alone | | Trial retry | 7 days after failure; at most 2 attempts in 30 days | Limits learning the planted set by repeated guessing | | Ongoing planted work | 1 synthetic job per 20 real jobs; same paid workflow | Tests drift after admission; costed in fees below | | Commit/reveal | Commit within 24 hours of assignment; reveal within 6 hours after both commitments | Enough time for own-machine work; commitments bind job ID, chain, block hash, schema, evidence hash and secret salt | | Challenge window | 7 full days after publication and evidence availability | Weekend and independent reproduction coverage; the clock does not start while evidence is inaccessible | | Liability release | After the window if unchallenged; otherwise after final resolution plus 3 days | No early recycling of disputed collateral | | Unbonding | 14 days from request AND all outstanding liability release conditions satisfied | Stops new assignments immediately and allows discovery of recently published mistakes | | Dispute timeout | 30 days from filing, then neutral closure, service refund and no factual slash | Avoids an administrator being able to hold an operator hostage forever | | Bond review | Monthly; require top-up to $100 equivalent only for future assignments | Avoids automatic liquidation and retroactive changes to accepted work | **[P]** Do not slash for a market move, a missed assignment, a factual disagreement or a correctly qualified unknown. A missed deadline loses that task's fee and causes reassignment. A proven wrong slashable statement consumes that arm's reserved $250 and 25% of its CLAUS admission units. A second proven error within 90 days, or proven fabricated evidence, consumes the remaining admission bond and suspends admission pending a fresh trial. Refunds are paid once per job, not once per mistaken field or arm. The 25% first penalty preserves a route back after an isolated error; repetition removes that concession. These are pilot deterrence budgets, not empirically optimized penalty rates. **[I]** A CLAUS-only operator bears work risk, collateral volatility and the issuer's administrative risk together. Falling collateral coverage can force top-ups exactly when refunds are needed; increasing its required token quantity can lock out capable operators. A holder controller with many wallets can still buy many apparent identities. A bond prices that attack; it does not eliminate it. **[P]** Offset capital and compute costs through paid assignments, published queue depth, prepaid customer orders and shorter release of undisputed job collateral—not emissions or idle-bond yield. Permit equipment or fixed-fee operating sponsorship with public controller attribution; do not create a protocol share of rewards for passive backers. If stable collateral is rejected, limit service refunds to recoverable CLAUS units and remove the dollar “made whole” language. I would not launch dollar guarantees against CLAUS alone. ## 4. C — Precisely what can be challenged and slashed **[P]** A control report must pin chain ID, block number **and hash**, contract address, runtime code hash, proxy implementation/admin slots, authority graph, exact call/storage evidence and uncertainty. Slashable examples are a wrong observed `owner()` address; claiming no upgrade path while documenting a reachable, enabled upgrade function; claiming minting is unavailable when a demonstrated authorized mint path exists at the pinned state; or reporting an LP position as locked when its stated custodian can withdraw it under the published criteria. A challenge must prove a false statement under the agreed schema, not merely a later change. **[P]** Exclude “safe,” “no rug,” beneficial-owner identity, honest intent, future governance behavior and absence of every possible exploit. “No mint authority found within these inspected paths” must list the paths and unresolved dependencies. An unknown ends the affirmative certificate for that field; it does not become a free, positive verdict. Repeated unproductive unknowns reduce assignments rather than trigger factual slashing. **[P — process and money]** 1. Anyone may file public evidence free. A deterministic replay can correct a report without convening a panel. To demand contested arbitration, post **$75**: $25 challenge stake plus a $50 first-panel expense advance. A customer or independent sponsor may fund it. No requirement to own a house or seat. 2. Select **5 conflict-screened controllers, 4 matching votes required**, paid $10 each for evidence-based, bounded first review. Collect evidence for 72 hours and decide within the next 72 hours. Publish independent reasons. No quorum means unresolved, never “the old answer wins.” This is my proposed configuration, not verification of the existing Claus panel. 3. For a successful challenge, return the $75 advance and pay the customer its actual checking fee, then one $25 bounty per job, then reimburse actual panel costs up to $50 from the slashed pool. The $50 panel allocation repays the challenger's spent expense advance; it is not a second panel payment in addition to that refund. The stable remainder goes to the reserve until its funding gate is met; only then is it converted to CLAUS and burned. Slashed admission CLAUS follows the same reserve-before-burn rule. This deliberately changes “all the rest burns.” 4. If only one arm is wrong and the fee was $65, the $250 stable slash is **$65 customer, $25 challenger, $50 panel, $110 remainder**. If both are wrong, both reserved tranches slash but the customer, bounty and panel are not paid twice. Already paid refunds or panel expenses cannot be claimed again. 5. A losing challenge pays the $50 panel expense and loses its $25 stake to the reserve. Duplicates on the same fact share one case; novel evidence can reopen it within the original window. No bounty is paid from new issuance or the reserve merely for opening a case. 6. Permit one appeal within **72 hours**, to **3 independent technical human arbiters, 2 agreeing**, with **$300** prepaid costs ($100 each). The losing party bears the appeal cost; return unused amounts. A sponsor can fund an evidence-rich appeal. An appeal may overturn the slash and restore collateral from the reserve. The report must remain marked disputed until finality. **[P]** Where IMD built the contract, no IMD seat, IMD builder/reviewer, Claus employee or party-affiliated controller may sit on the first panel. Use five outside operators on the same disclosed fee schedule and the independent technical appeal panel above. Choose and publish the roster, replacements, conflicts policy and settlement-signature mechanism **before accepting the job**. If no qualified unconflicted panel is available at the budget, quote the real higher fee or decline binding service. “Outside arbiter” is not a sufficient specification, and an NFT balance is not a qualification. **[I/P]** Collusive customer–arm–challenger rings must lose more than they can extract: no alleged project losses are reimbursed, customer refund cannot exceed its paid fee, the $25 bounty is smaller than the $250 forfeiture, and no reserve-funded bonus supplements it. Require evidence commitments before revealing counterpart reports, randomly assign pairs, reject known controller overlap and keep one case per fact. These controls make self-challenge farming costly; they cannot prove absence of secret collusion. Freeze only the disputed job's tranche, not every unrelated job of a challenged operator. ## 5. D — Fee flow that pays for the second check **[P]** Reserve every component when the customer prepays. “Arms first” should describe payment protection, not permission to exhaust a job's dispute budget. Pay each arm for an independently evidenced submission, including a useful unknown; do not pay only for matching another answer. Mismatches trigger a paid third investigation or an unresolved report, never iterative copying until two answers agree. **[P — control-check budget, 20 retail jobs/month]** | Component | Dollars per check | Cost assumption | |---|---:|---| | First arm | 15 | 20 minutes operator attention at $30/hour = $10; inference/retries $3; machine/RPC allocation $2 | | Independent second arm | 15 | Same budget; independence must actually be purchased | | Planted-defect programme | 2 | A $40 synthetic exercise every 20 real checks, including preparation and checking | | Disputes/refunds reserve | 8 | Under a pilot assumption of 2% first-panel disputes, 1% $65 refunds and 0.5% $300 appeals, expected cost is $3.15/check; remainder funds uncertainty and replenishment | | Coordination | 20 | $5 variable assignment/publication plus $300 monthly fixed cost / 20 checks | | Surplus | 5 | Initially retained; potentially burnable only after funding gates | | **Retail quote** | **65** | **All amounts proposed; no measured production cost is claimed** | **[I/P]** The corresponding quote is **$50 + $300/N** for N prepaid checks per month under those assumptions: $65 at N=20, $55 at N=60, $53 at N=100. At zero orders there is no fee-funded business. The $50 figure is a limiting budget, not a sustainable low-volume price. If measured operator time doubles, the two attention budgets add another $20. Reprice rather than hide unpaid operator labour. Stable collateral is returned capital and is not counted as revenue. **[P]** Seed the reserve with **$2,000** of explicitly authorized, existing funds or sponsorship before promising refunds. Keep stable refund/dispute balances segregated from working CLAUS inventory. Start burning only after at least **100 completed paid checks and 90 days** of operation, no unpaid operator invoices or overdue claims, and reserve coverage of the greater of **$2,000** or **all outstanding maximum service refunds + 90 days' stressed dispute expense**. Stress dispute incidence at **10%**, not the 2% budget expectation. Review funding monthly. The burn amount can be zero indefinitely. **[P]** Publish quarterly actual costs and related-party coordination payments. Reimburse a disclosed budget at cost; do not give the coordinator an unlimited claim called “cost.” The existing 1% pool treasury fee remains a separate current policy unless Claus explicitly changes it. The new service has no percentage treasury allocation and no NFT allocation. **[P — product sequence]** Sell the reproducible control form first. Add upgrade reviews at **$250 minimum**: $140 for two arms, $20 replay/testing, $40 reserve, $30 coordination and $20 provisional surplus. Require the exact implementation/calldata/code hashes to be frozen and at least a **14-day** upgrade timelock. A timelock alone does not make a Claus verdict binding: actual blocking requires a project-installed execution gate recognizing the reviewed payload and unresolved disputes. Without that gate, sell an advisory review only. **[P]** Add watch at **$30/contract/month** for deterministic change detection and publication: $12 collection/monitoring work split between two independent watchers, $8 reserve, $5 coordination, $5 provisional surplus. Target a 5-minute alert within a measured 95% service level, explicitly excluding unavailable chain data; retain outage records. Mark a single watcher's alert unconfirmed until the other reproduces its block evidence. Semantic recertification is a separately approved $65 paired check, not hidden unlimited work. Renew collateral and scope monthly. Do not market these budgets as full smart-contract audits. ## 6. E — Token use, supply arithmetic and issuer controls **[R — supply source]** The accessible Codex Scan token page served **983,872,663.4004861 CLAUS** as “max total supply,” **984,256,771.92** as circulating, 18 decimals and 3,983 holders when read during this session. The page does not bind those fields to a common block. The circulating value exceeding its total is a warning against treating the page as a reconciled supply proof. A Blockscout search extract separately displayed 984,017,154.588; its live page yielded no readable data. I use the former total only as an explicitly provisional denominator. [Codex token page](https://scan.codex.io/robinhood/address/0x163015276f3244bc2ccb78b4d69cc8a4069effac?tab=holders), [Blockscout token page](https://robinhoodchain.blockscout.com/token/0x163015276f3244bc2ccb78b4d69cc8a4069effac). **[I — calculation, not a live settlement quote]** Let q be the entry-date CLAUS units needed for the recommended $100 admission bond and S the contemporaneous onchain `totalSupply()`. The exact formulas are **nq CLAUS** and **100nq/S percent**. Solely translating the explorer's displayed accounting conversion gives q ≈ **159,286.397 CLAUS**. The following is therefore an indicative admission scenario, not a claim that this quantity is already bonded or an executable quote; transaction costs and slippage are excluded. No forecast is involved. | Arms, each separately controlled | Admission CLAUS, approximately | Share of provisional S | Additional stable collateral if each has 2 open control checks | |---:|---:|---:|---:| | 10 | 1,592,864 | 0.1619% | $5,000 | | 100 | 15,928,640 | 1.6190% | $50,000 | | 1,000 | 159,286,397 | 16.1897% | $500,000 | **[I]** In the otherwise comparable **CLAUS-only** alternative, $100 admission plus two $250 job tranches would require six times these token quantities at that conversion: roughly **95.7458%** of the displayed supply for 1,000 arms before other uses. This is a capacity stress test, not a prediction of adoption or a justification for raising the bond to absorb supply. It is a reason to separate dollar remedies from the CLAUS admission stake. Actual escrow demand depends on concurrency and challenge retention, not registrations alone. **[P]** Entry conversion should use an independently reproducible executable route and a 30-minute observation, reject data older than 15 minutes and stop admission if a $100 conversion would move the route by over 1%. Do not automatically trade an operator's collateral to maintain a target. Where a credible quote cannot be obtained, keep existing units and stop new exposure. These are manipulation/availability gates, not assertions that a suitable deployed oracle already exists. **[I]** Unlimited owner minting can dilute penalties and counterfeit the apparent cost of admission. Transfer blocking can stop bond return, refunds or burning. A proxy upgrade can recreate either power after a nominal role revocation. A founder-controlled authority contract is not independent merely because its immediate owner is another contract. The chain of ultimate authority, not the first `owner()` result, matters. **[P — what Claus must lock]** Publish the proxy, implementation, admin/beacon, governance, delegated signers, fee setters and LP custodians at one finalized block. For the bonded token, remove discretionary extra minting and any seizure, blacklist, transfer-tax or transfer-pause path, and prevent their reintroduction by upgrades. Either freeze that implementation after migration reconciliation or enforce those invariants through an independently reviewed, non-bypassable authority design. An announcement is insufficient. **[P]** Where upgrades remain necessary in coordination or hooks, use a **3-of-5 multisig** of genuinely separate controllers and a **7-day onchain timelock**; publish signers and all role-grant/recovery paths. Do not permit it to rewrite existing bond terms, take deposited principal or halt matured withdrawals. Bond and refund escrows should be immutable by version, with voluntary migration only after liabilities settle. Limit an emergency switch to **48 hours of stopping new jobs**, automatically expiring; it must not freeze withdrawals. Lock project-owned LP principal for at least **180 days** if its continued availability is used in admission conversion assumptions, and disclose that third-party LPs remain free to leave. A reversible mint-role revocation is not a supply lock. ## 7. F — Demand measured separately from willingness to pay **[R — repository observation]** Anonymous GitHub metadata for `identity-md-launches`, sorted by creation time descending, pages 1–2 at 100 entries/page, was read at **2026-11-12 11:38:09–10 UTC**. Filtering `created_at` to the seven complete UTC days November 5–11 gives the following. All 120 counted entries were non-forks and nonempty by GitHub's `size` field. [Public page 1](https://api.github.com/orgs/identity-md-launches/repos?sort=created&direction=desc&per_page=100&page=1), [page 2](https://api.github.com/orgs/identity-md-launches/repos?sort=created&direction=desc&per_page=100&page=2), [public organization](https://github.com/identity-md-launches). **[I — counts calculated from `created_at`, not updates]** | UTC creation day | Public repositories | |---|---:| | 2026-11-05 | 20 | | 2026-11-06 | 40 | | 2026-11-07 | 22 | | 2026-11-08 | 18 | | 2026-11-09 | 2 | | 2026-11-10 | 10 | | 2026-11-11 | 8 | | **Total / daily mean** | **120 / 17.14** | **[I — limits]** This is **repository creation**, not 17.14 deployed tokens or paying customers per day. Workflow stages, websites and research are included; public repositories can be created before delivery, and multiple stages can belong to one project. For example November 11's eight include both `launch-3206-workflow-contract-stage-context` and `launch-3213-workflow-frontend-stage-context`. The first page contained this research assignment's empty repository metadata; it was outside the complete-day count and its contents were not read. Private or unpublished work is excluded. No contract-deployment rate is established by this count. **[R/I — Robinhood explorer volume]** Codex Scan's launchpad explorer displayed **1.28K tokens created on covered launchpads in the last 24 hours**, with rows Pons 929, LONG 238, Flap 67, Bankr 16, Clanker V4 14, UniswapCCA 11, Virtuals 5 and Flywheel 1. Their sum is **1,281**. The same panel displayed 7 graduations. This is a rolling explorer observation on November 12, not a complete UTC calendar-day chain census; the panel did not expose an exact interval-end timestamp or block. It excludes untracked factories/direct deployment and does not count independent customers. [Launchpad explorer](https://scan.codex.io/robinhood/launchpads). **[P — first buyers and prices]** Approach deployers needing an authority disclosure, launch factories or launchpad frontends that can buy repeated standardized checks, and wallet/explorer teams that need attributable control fields. IMD is a plausible purchaser of independent checks on its deployments but must not supply its own deciding panel. Sell $65 retail checks, $55 prepaid checks only at a real 60-check monthly commitment, $250-minimum upgrades and $30/month watches with explicitly priced recertification. A $5 consumer badge cannot support this paired process. **[I/P — commercial test]** Neither 120 repositories nor 1,281 launches proves anyone will pay. A six-hour or one-day manual review cannot protect a token that exists for minutes. Start with persistent projects and repeat distribution customers. Require **20 prepaid checks from at least 5 unrelated buyers within 30 days**, publish actual cost and unknown/dispute rates, and stop expanding admission if that test fails. At the proposed attention budget, 20 checks buy only about 13.3 operator-hours across both arms, insufficient to employ a large swarm. Do not recruit 1,000 idle arms on the strength of gross chain activity. ## 8. H — Concentration: addresses, method and block ### Ethereum: a reproducible pinned reconstruction **[R — contracts and block]** Claus NFT: `0x80396c7131159eB92E7E84839e51C9887A7A95d9`; Claus NFT staking: `0x4e5916428452891E1deb59d162c20266ED08CAD4`; IMD seats: `0x0000eC93127BAA929E58E97dd0095A2BFb38ec1D`. The NFT address is published by Claus, the seat address by IMD, and Routescan's verified staking source names `ClausNftStaking` with a `stakerOf` mapping and indexed `Staked`/`Unstaked` events. [Claus collection](https://claus.si/read/Hooks/NFTs), [IMD contract listing](https://imd.fun/docs/api/), [staking verified-source response](https://api.routescan.io/v2/network/mainnet/evm/1/etherscan/api?module=contract&action=getsourcecode&address=0x4e5916428452891e1deb59d162c20266ed08cad4). **[R]** Ethereum end block **26,398,410** (`0x192ceca`), hash **`0x1282873752172627b577ba44dd7fe817fae93131243ecbabe5f96774bfd995a0`**, timestamp **2026-11-12 10:57:59 UTC**. [Pinned block response](https://api.routescan.io/v2/network/mainnet/evm/1/etherscan/api?module=proxy&action=eth_getBlockByNumber&tag=0x192ceca&boolean=false). **[I — method]** Fetch all ERC-721 `Transfer` events from block 0 through that block for each collection using the public [Routescan logs API](https://docs.routescan.io/api/etherscan/logs), topic `0xddf252ad1be2c89b69c2b068fc378daa952ba7f163c4a11628f55a4df523b3ef`. Page in 1,000-record batches. At the API's 10,000-result window, restart inclusively at the last returned block and deduplicate transaction hash/log index. Replay in block/log order, asserting that each non-mint transfer's sender owns the token. Burned tokens leave the count. Sum surviving tokens by address. **[I — staking look-through]** Replay 278 `Staked(address,uint256,uint256)` and 93 `Unstaked(address,uint256,uint256)` records for the staking address at the same end block. Replace the staking custodian with each token's active staker, then aggregate that address's directly held NFTs too. This resolves **185 staked NFTs**, not the ultimate humans or entities behind their addresses. Other undisclosed custody/control relationships remain unresolved. **[I — repaired source anomaly]** The IMD log response contained the same transfer twice with different log indices for transaction `0xdc2330feeb315bae1911579fcc52554c0947ba0baf18ad16651477b964be1adb`. Its [transaction receipt](https://api.routescan.io/v2/network/mainnet/evm/1/etherscan/api?module=proxy&action=eth_getTransactionReceipt&txhash=0xdc2330feeb315bae1911579fcc52554c0947ba0baf18ad16651477b964be1adb) contained one relevant transfer at log index `0xc7`. Replacing that transaction's indexed records with its receipt removed the impossible double-send. The final replay used **1,318 Claus transfers and 11,130 IMD transfers**, conserved ownership throughout and ended with 300 and 2,000 NFTs. This check caught an index error; it is not an independent full-node verification of the index. **[I — measured results at that Ethereum block]** HHI below is `10,000 × sum((balance/collection_total)^2)`; it measures address concentration, not controller concentration. | Collection and treatment | Addresses | Largest address | Largest 10 | Largest 20 | HHI | |---|---:|---:|---:|---:|---:| | Claus, raw custody | 69 | 185 / 61.67% | 232 / 77.33% | 251 / 83.67% | 3,871.56 | | Claus, staking attributed to stakers | 147 | 22 / 7.33% | 116 / 38.67% | 152 / 50.67% | 216.89 | | IMD, raw seat ownership | 789 | 77 / 3.85% | 385 / 19.25% | 594 / 29.70% | 69.405 | **[I — largest 10 addresses; Claus column uses staking look-through]** | Claus address | NFTs | IMD address | Seats | |---|---:|---|---:| | `0xc4210bfd73206c3208ae1f0854ee146ed8f1271b` | 22 | `0xb7a832db45f240f876432bf8b222bff30b4be8ff` | 77 | | `0x85b1f16a7cc5da288dc796ae8300099c3ac9c66c` | 16 | `0xb64199cd690aa4e78f1a6625660347ce63031d72` | 54 | | `0x0feaf114f65361dfaaf981f2662e3843376dacbe` | 16 | `0x80271ce20184e38f4afe90d4ca134304d197aca2` | 47 | | `0xdb585820d3b56dbe452b2d7774e7852dbd0976ea` | 15 | `0xe3de112ae639b45b7d22c4e42742020db679d80c` | 34 | | `0x9bf5c4f08c797a05933fcd3db4cb94b62b184709` | 11 | `0x6ba9ad599e6271252817933831e3148936cf742a` | 30 | | `0x28c1c00e6e27a2a9a93c1bbea01396a372ab1a2d` | 11 | `0xabe0cdd036456fcc1b64d1ddb7aa3bdf7d5b98b1` | 30 | | `0x38d7274dc48a106320bfc023bef09ac4de7de13c` | 8 | `0xd4bb7351a9b0d1b236e7523da7a972db4a100f08` | 30 | | `0xff958cb19c81532ca1e6521f1208ab2c0c5ca4a6` | 7 | `0xf98c3bf5f3544f64e60560304b58dca900a3c4db` | 28 | | `0x7665c425b5f3a0249644024bb32a179ab2d5e0a2` | 5 | `0xb19889f4b6976beba437d539eee566dc31805dc2` | 28 | | `0x76f91dcf2aa8f542a5baf8ac9b06b074d80d4015` | 5 | `0x7cc6f600b21f6fe03daa9da6ad6dc416c2d6fc4e` | 27 | **[I — limits]** Ties at the tenth position can substitute another equally sized address without changing the concentration total. These are not 147 or 789 proven independent people. The brief's 148 NFT payout recipients concern an earlier snapshot; 147 current staker-adjusted addresses at a later block is not evidence that its payout count was wrong. Raw custody would misleadingly present the staking contract as a 61.67% whale. IMD beneficial ownership behind custodians, lending and common controllers was not resolved. ### Robinhood CLAUS: useful rows, no defensible pinned distribution yet **[R]** The public [Codex holder page](https://scan.codex.io/robinhood/address/0x163015276f3244bc2ccb78b4d69cc8a4069effac?tab=holders) served the following first ten balances, rounded here to whole CLAUS. Its HTML data retained 18-decimal balances. Observation was during November 12 research; **the holder dataset's block is not supplied**. The latest trade block on the page, 112,491,498, is not a valid substitute for a holder snapshot block. | Rank | Address | CLAUS, rounded | |---:|---|---:| | 1 | `0x8366a39cc670b4001a1121b8f6a443a643e40951` | 96,828,661 | | 2 | `0x63b9a33525ebbfc765ac9c81e852c57bff0a2cef` | 40,309,610 | | 3 | `0x44e4208f7806e009beff792f7c323ff1b359d308` | 27,029,617 | | 4 | `0x1c57ac0d42d70665bbcba2f11e58cd9210500fb9` | 24,924,923 | | 5 | `0x9d42cee9a2a6bd1b8a185bb9aed6dcb5ce53b7ed` | 23,399,613 | | 6 | `0xabb0d66ec158a47a5bb0a4439bf1f148bd2d3aa6` | 18,374,327 | | 7 | `0x27dace8fb231ba6c1121a1f96b0ee815f8cf26e5` | 17,184,445 | | 8 | `0x1ab48085ac8f27403e344a52e6a18b8c78b2c308` | 15,723,204 | | 9 | `0x77b6e288b1e578ebbcd5708d48e4fb929bc6a44d` | 15,667,440 | | 10 | `0x135413bb6883d648c4faf261e7d3a9e6ede740a1` | 14,577,302 | **[I]** Summing unrounded rows gives **294,019,141.727940287462266031 CLAUS**, or **29.8839%** of the page's total-supply field. The first address alone is **9.8416%**. The page's headline instead says **21.3409%** for its top ten. These are not reconciled metrics, and I do not choose one as the definitive current concentration. **[R/I]** The [first address page](https://scan.codex.io/robinhood/address/0x8366a39cc670b4001a1121b8f6a443a643e40951) identifies an Uniswap V4 exchange contract. The [second address page](https://scan.codex.io/robinhood/address/0x63b9a33525ebbfc765ac9c81e852c57bff0a2cef) identifies a beacon-proxy contract, labelled Share Flaunch, with unverified source. Counting either as an ordinary independent holder would be unjustified. Excluding them without attributing their beneficiaries would also be unjustified. The observed raw totals are useful operational custody measurements, not controller voting shares. **[U/P — exact missing measurement]** A block-pinned Robinhood holder count, controller-adjusted top ten, HHI and supply reconciliation remain unavailable. To complete them, pin a finalized Robinhood block/hash, replay every CLAUS transfer from deployment to that block, reconcile mint/burn events with `totalSupply`, and read candidate custodian, reserve, proxy/admin and LP-position records at the same block. Report both raw custody and resolved beneficial balances; publish unresolved contract balances separately. Do not add old Ethereum balances to Robinhood supply. No founder-only secret is needed to perform this measurement once usable public chain access is available. ### One controller, one voice **[I]** It cannot be enforced cryptographically against a controller using fresh wallets, independent funding routes and different machines. Token and NFT concentration are lower bounds on common-control concentration. Different signatures, model names or IP addresses are not proof of independent decision-makers. ERC-8004 supplies a public identity and reputation format, not personhood. [ERC-8004 security considerations](https://eips.ethereum.org/EIPS/eip-8004). **[P — cheapest workable test]** Use a signed controller/conflict declaration, public payout/controller keys, the same 20-fixture trial and a seven-day probation. Refuse a pair if keys, declared controller, direct funding origin or prior signed admissions establish overlap; treat common exchange funding or shared hosting only as a review flag. Pair randomly across declared controllers and prefer different model/provider stacks where available. Audit 5% of completed pairs for duplicated evidence trails or impossible independent timing. Publish exclusions and allow evidence-based appeal. Call this **screened independence**, not enforceable one-human-one-vote. Do not slash solely from a clustering heuristic or require invasive personal identification for a low-value control form. ## 9. G — The strongest NFT objection and the honest answer **[R]** The October 6 Claus journal did more than describe artwork: it recorded fee rewards and an intention to add future features and privileges. It is understandable that a holder would have relied on that direction. The current NFT read page also preserves already earned rewards while saying new Ethereum fee allocations are paused. [First NFT payday](https://claus.si/read/Journal/journal:claus-nft-rewards-first-period-20261007), [current NFT read page](https://claus.si/read/Hooks/NFTs). **[I — strongest objection]** “We supplied early capital and attention under a story of increasing utility; now a new product uses that community while refusing the houses any participation.” That is stronger than simply demanding passive income. The token migration payment should not be presented as the holder having agreed to every future change. **[P — honest answer]** A new checking business must pay the people doing independent checking, its customers' remedies and its actual costs. It cannot promise permanent income to a prior collectible without either raising customer fees or underpaying work. Preserve what is actually owed and existing nonfinancial access; acknowledge that future privileges were discussed but this particular new income stream was not agreed. Publish a rights-and-payments reconciliation, with no claim that a website announcement cancels an accrued contractual entitlement. **[P — courtesy]** Offer an archival acknowledgement, an optional link from each existing house to its holder's project, and one public walkthrough of operating an arm. Make the operating documentation available to everyone at the same time. No fee share, exclusive customer queue, trial waiver, discounted bond, new compute subsidy or special burn distribution. Courtesy should not become a continuing economic preference. ## 10. What I would cut or change plainly **[P]** - Cut the **CLAUS-only dollar guarantee**. Retain CLAUS admission and use stable job collateral; otherwise promise only token-denominated, limited remedies. - Cut **“customer made whole”** if it implies reimbursing funds lost by using the reviewed project. Promise the paid checking fee back, with stated caps and exclusions. - Cut **IMD seats automatically skipping the checking trial**. Building reputation can shorten onboarding paperwork; everyone must pass the same authority fixtures. - Cut **matching answers as the acceptance criterion**. Require matching reproducible evidence; pay for useful disagreement and route it to resolution. - Cut **automatic slash-by-majority**, unlimited liability and indefinite dispute locks. - Cut **burning before reserve and fixed costs are funded**, and any unlimited coordinator expense privilege. - Cut **house-based seats, launch allocations and payment in the reviewed token**. Require declarations of holdings and control before assignment and no trading/accepting that token from assignment through the challenge window; do not pretend undisclosed holdings can be perfectly detected. - Cut **a binding upgrade badge based solely on a timelock**. Binding enforcement needs an actual execution gate and a frozen payload. - Keep **no numerical cap on admitted arms**, but make clear that admission gives no work entitlement. Exposure limits, pair independence and demonstrated demand determine assignments. ## 11. Questions only the founders can settle **[U — decisions and commitments, not research gaps disguised as facts]** 1. Will Claus accept stable per-job collateral alongside a CLAUS admission bond, or insist on token-only remedies with explicit dollar-coverage limits? Which issuer, chain and verified settlement-token contract will it accept? No token address or stable-asset availability is assumed here; stable collateral still carries its own issuer and redemption dependencies. 2. Will the founder irrevocably constrain minting, upgrade, recovery and transfer-control authority as above, and which actual signers will hold the remaining timed powers? Public addresses and current storage are researchable; willingness to surrender control is a founder decision. 3. Which existing NFT redemption, reward and sponsored-inference obligations does Claus acknowledge, and what exact migration action changed each one? Will disputed obligations be funded before a new service starts? 4. Who will supply the $2,000 reserve seed and the $300 monthly fixed coordination budget if prepaid demand is insufficient? No unspecified treasury entitlement is assumed here. 5. Will IMD accept paid outside checking and fund an initial customer batch? Will its workers disclose conflicts and decline launch-token allocations on projects they review for Claus? 6. Who are the actual independent non-IMD technical arbiters and alternates, will they accept the fee schedules, and who will implement and obey their decisions? No named person's participation is assumed. 7. Will both founders accept the limited service-fee remedy, public uncertainty labels, public failed-trial statistics, controller declarations and a stop decision after a failed paid-demand test? ## 12. I — Five sentences for a CLAUS holder **[P — proposed holder-facing explanation; five sentences]** Anyone can operate a Claus arm by posting a returnable CLAUS bond, passing practical checks and supplying collateral for the jobs they take. Two independently assigned arms must publish evidence for the same contract facts, and a proven false statement can cost the responsible arm its collateral. Customers pay for work, with fees funding the arms, challenges, refunds and disclosed operating costs before any surplus burns CLAUS. Holding CLAUS alone earns no checking fee, and the 300 NFTs receive no new income or preferred access from this service. The token's administrative powers and the dispute rules must be locked down before Claus can credibly ask anyone to trust the bond. ## Reproduction and local checking record **[R/I]** The public Ethereum log requests use the following URL pattern, substituting the three published contract addresses and the appropriate event topic; no API key is required. The first Claus transfer page is [directly readable here](https://api.routescan.io/v2/network/mainnet/evm/1/etherscan/api?module=logs&action=getLogs&address=0x80396c7131159eb92e7e84839e51c9887a7a95d9&fromBlock=0&toBlock=26398410&topic0=0xddf252ad1be2c89b69c2b068fc378daa952ba7f163c4a11628f55a4df523b3ef&page=1&offset=1000). ```text GET https://api.routescan.io/v2/network/mainnet/evm/1/etherscan/api ?module=logs&action=getLogs&address=COLLECTION &fromBlock=START&toBlock=26398410&topic0=EVENT&page=PAGE&offset=1000 Staked topic: 0x1449c6dd7851abc30abf37f57715f492010519147cc2652fbc38202c18a6ee90 Unstaked topic: 0x7fc4727e062e336010f2c282598ef5f14facb3de68cf8195c2f23e1454b2b74e ``` **[I]** Local checks asserted event uniqueness after pagination, end-block bounds, correct event topics, valid ownership at every transfer, staking custody matching the active staking ledger, and final supplies of 300/2,000. Concentration uses exact integer counts before rounding. GitHub counts use creation dates, seven complete UTC days and no repository-content inspection. CLAUS concentration was recalculated from exact integer balances rather than copying its inconsistent headline. These are reproducibility checks by this author, not an independent audit. **[R — provenance fingerprints]** SHA-256 of the two raw GitHub response bodies at the stated observation time: page 1 `c1e7e9758506e7685c4c2f631dcf7bc4e8c1524fad34a9105e5f3bc726a892ae`; page 2 `40bdb9418f0b87219801a79e3ebff0fe38ecaa10e1175dba80ebdfcf77a57dbc`. These identify changing source observations; the report does not require a verifier to have network access or claim that a digest alone proves a source's truth.