# Claus arms: a CLAUS work bond as the seat. Final report, combining five independent views Date: 12 November 2026. This is a proposal for discussion. Neither Claus nor IMD has agreed to any of it. It contains no price predictions and no investment language. Every dollar figure is a proposed service price or cost assumption, not a reading of the market. ## How to read this report This report brings together five views (V1–V5). Each was written by a researcher working alone from public sources. The views are in `artifacts/view-1.md` … `view-5.md`, and their SHA-256 hashes are listed in the last section. Every statement carries one of these labels: - **[R: Vn]** read from a public source or a public read-only chain call by view *n*. The source is given in that view and is repeated here where it matters. - **[I]** an inference made in this report from [R] material. - **[P]** a proposal. The views that make it are listed as "n/5: Vx, Vy". - **[U]** unverified, conflicting, or not measured. - **[X]** a spot check made for this report: a read-only public RPC call to `https://rpc.mainnet.chain.robinhood.com` at Robinhood Chain block **112,514,839**. Several views cite outside networks (The Graph, UMA, Kleros, Truebit, Arbitrum, Chainlink, EigenLayer) from memory and say they did not re-fetch them. V1 marks these [R-recalled] and V5 marks them [U]. This report keeps those caveats. Where V2, V3 or V4 fetched the primary page, that is said. **Tally rule.** A number given for a parameter lists every view that proposes the same value or the same rule. Views that propose something different are listed beside it. A point backed by only one view is kept when it is well sourced. --- ## 1. Corrected facts These come first, as the brief asks. The brief's context is shown in quotation marks. | # | Context says | Corrected or qualified fact | Status and views | |---|---|---|---| | F1 | CLAUS moved to Robinhood Chain, token `0x1630…fFAC` | **Correct.** Chain ID 4663, symbol CLAUS, 18 decimals. The Ethereum token `0x1b54E762aa34CF6E28E9C082F2848e28E45DA6b8` is "retired" for migration but **still exists and can still be transferred**. | [R: V1, V2, V3, V4, V5] (claus.si, claus.si/docs/migration). Ethereum token still transferable: [R: V1, V3] | | F2 | 2% pool fee, half buyback and burn, half treasury | **Correct.** 1% buys back and burns CLAUS; 1% goes to the CLAUS DEV treasury in ETH. **Qualification:** the live operations status splits the treasury half into `founderPips=9000` and `advisorPips=1000`, with advisor address `0x1C31B012a8565567A72F7c661D7f0d9f92182853`. So "half treasury" is really a founder share plus an advisor share. Automatic liquidity additions have stopped (V5). The legacy Ethereum pool charges 2.3% (V1). | Split: [R: V1, V2, V3, V4, V5] (claus.si/Hooks, claus.si/read/Hooks/Funding). Founder/advisor subdivision: [R: V2 only] (https://claus.si/api/migration/status) | | F3 | Each of the 300 NFTs was paid 50,000 CLAUS, 15M to 148 wallets | **Correct.** The payment came from the existing reserve, **not from new minting**. Staked NFTs were credited to their stakers at Ethereum snapshot block 26,357,823. The payment transaction is `0xbd404685…9046`. It was a **separate** payment from the 3,805 direct holder allocations. | [R: V1, V2, V3, V4, V5]. Snapshot block: V4, V5. Transaction: V2, V4. 3,805 allocations: V1, V3 | | F4 | "The token is owned by a contract owned by one plain wallet" | **True but incomplete, in a way that matters.** (a) `owner()` = contract `0x979D5810D32292358639f25294472f01c95bDc46`. That contract's `owner()` = `0xBb145cA83272c3806D4dDc75cA1D5514789cF1C5`, which has no code (an EOA). (b) **The token is an upgradeable ERC‑1967 proxy.** Its admin slot holds `0xcbc4bbd337e62c926c995e66a231e5e2b0a722a8`, and that address's `owner()` is the same `0x979D…` contract. Its implementation slot holds `0xd68727ddd50e8ba07e4be20d37a5ecf3dd814bff`. So **the one EOA can replace the token's code.** (c) The administrator contract also has a delegated `operator()` = `0x8aca8890d34086d7a30b9c06fae25e0c3cda4cfa`, which can call managed targets, the token ProxyAdmin among them. | Owner chain: [R: V1, V2, V3, V5]; V4 did not verify it. Proxy admin: [R: V1, V2, V5]. Admin owned by `0x979D…`: [R: V2, V5]; V1 found the admin but did not read its owner. Operator: [R: V2 only] (Sourcify-matched source). [X]: the admin slot, implementation slot and `owner()` all match at block 112,514,839 | | F5 | (mint authority, pause) | **Minting is live and has no amount limit.** The implementation has `mint(address,uint256)` restricted to the owner. V5 simulated `mint` from `0x979D…` and it succeeded; from any other address it reverted. Claus's own contract page says the developer wallet controls minting, fees and project-owned liquidity, with no enforced mint cap or delay (V3, V4). **No ordinary transfer pause or blacklist** was found in the current implementation. `setMigrationPaused` only gates migration receipts. Any of these powers could be added by an upgrade. **Conflict:** `claus.si/build.txt` says core upgrades are unavailable (V4). V2 shows that the matching `hook-health.json` refers to chain 1 (Ethereum), not 4663. That statement must not be applied to Robinhood CLAUS. | Mint: [R: V2, V4, V5] plus Claus disclosure [R: V3]. V1 left it [U]. No pause: [R: V2, V4, V5]. build.txt conflict: [R: V4], explained by [R: V2] | | F6 | (supply) | Total supply was **983,872,663.400486 CLAUS** at blocks 112,496,202–112,498,036. Replaying every Transfer log (14,877–14,881 logs from block 107,148,941) gives the same sum. **The supply is not capped**, because `mint` is live. A [X] read at block 112,514,839 gives **983,788,670.50**, about 84,000 CLAUS less. That fits buyback burns reducing `totalSupply` (V5 found a zero balance at `0x…dEaD`). Any supply share must therefore be quoted with its block. | [R: V1, V2, V3, V5] by `totalSupply()`. Log replay matches: [R: V2, V3, V5]. V4 used an explorer figure and calls it provisional, because the same page shows circulating supply above total. Decline: [X] | | F7 | "Claus Jobs escrow runs on Ethereum with the old token" | **Correct for CLAUS-denominated deals, with a qualification.** Public deal terms name an immutable escrow `0xc2292af8b51109a27a4b4e28e9cf9f3aede4190c` on Ethereum, settling in **old CLAUS** `0x1b54…`, with arbiter `0xB104063f7a25a712a5899861941eb5cEB24EE105`. ETH deals use a different escrow. The Jobs board itself is a free task board, and paid agreements are a separate rail (V4). Settling in a retired token is a defect in its own right (V5). | [R: V2] (https://claus.si/__network/v1/public/chains/1/deals?asset=CLAUS). Qualification: [R: V4] (claus.si/build.txt). V1, V3 and V5 could not find the escrow address | | F8 | "Disputed Claus deals already go to a panel of five IMD members where four must agree" | **Partly correct.** Claus's public capabilities list arbitration as a **limited pilot**: five panel members, quorum four, case and principal limits, and an explicit statement that **no answer or payout is guaranteed**. That source does not show that the five are IMD members. IMD's own oracle sets panel size and quorum per question (its example is 5 of 5). A panel that disagrees gives no answer, and the fee is still spent. IMD charges 0.5 IMD per oracle request. | 5/4 pilot: [R: V2] (https://claus.si/__network/v1/public/capabilities). IMD oracle configuration: [R: V1, V2] (imd.fun/skill.md, imd.fun/docs/api). V3, V4 and V5 could not verify the 4-of-5 rule. "IMD members": [U] | | F9 | "Three earlier IMD swarm reviews cut houses-as-seats and burn-for-a-permanent-seat" | The public v0.3 page records that houses as seats and a flat CLAUS price were cut. **The burn-for-permanent-seat cut does not appear on that page.** The page also says "no burn in the pilot" and "reviewers are paid regardless of verdict". The tested design's slash and remainder burn are therefore **a change from the public version**, not a continuation of it. | [R: V1 only]. V2 and V4 could not open swarmcard.imd.site | | F10 | (implied) the 300 NFTs are "earning" | **Mostly no longer, but not zero.** New Ethereum fee allocations to NFTs are paused, and rewards already earned keep their terms (V1, V3, V4). **Conflict on inference sponsorship:** V5 reads `claus.si/docs/nft-inference` as saying NFT-sponsored inference has ended (`410 Gone`, `nft_sponsorship_retired`). V2 reads the same guide as describing up to $5 per NFT per day of sponsored inference, conditional on funding. *To settle this:* fetch that page and the inference endpoint and record the response. A third-party aggregator's claim of a 0.35% NFT share of volume could not be confirmed on any Claus page (V5). An Oct 7 2026 Claus journal entry recorded NFT fee rewards and an intention to add "future features and privileges" (V4). | [R: V1, V3, V4, V5]. Conflict V2 vs V5: [U]. Journal: [R: V4] | | F11 | IMD has 2,000 seats | **Correct.** Contract `0x0000ec93127baa929e58e97dd0095a2bfb38ec1d`, totalSupply 2,000. The docs say "free, one per wallet", but **this is not enforced at token level**: one address holds 77 seats. IMD gives **8% of each launch to connected seats and 2% to accepted contributors**. That clashes with rule 7 of the design (arms never hold what they check). | [R: V1, V2, V3, V4, V5]. Allocations: [R: V2, V3, V4] | | F12 | "No treasury cut" | True only **inside** the proposed fee waterfall. A customer who buys CLAUS in the official pool to pay a dollar quote pays the 2% swap fee, half of it to the treasury. So does an arm that sells its pay. | [I: V1, V2] from F2 | | F13 | The Claus NFT contract | The 300-NFT contract `0x80396c7131159eB92E7E84839e51C9887A7A95d9` on Ethereum has `owner()` = **the same EOA `0xBb14…`** that ultimately controls the token. | [R: V5 only] | | F14 | "Upside … shared by every holder alike" | A burn lowers supply. It credits no wallet and gives no holder equal cash. "Alike" holds per token, not per holder. | [I: V2] | --- ## 2. The design we recommend ### 2.1 Verdict **[P, 5/5: V1–V5]** Keep the core: an open, returnable CLAUS bond, a trial of planted defects, pay for work only, and sealed paired checks. It is the only one of the five seat models that puts the operator's own capital behind a wrong fact. The views agree just as widely on what must change before it is credible: 1. Slash only a **bounded, per-job tranche**, never the whole bond for one error (5/5). 2. Make slashable only **atomic facts that can be re-read at a pinned block** (5/5). 3. **Fund a reserve before any burn** (5/5). 4. **Keep IMD off any dispute about a contract IMD built** (5/5). 5. **Lock the token's mint and upgrade powers before taking anyone's bond** (5/5). 6. Stop claiming that "one controller, one voice" can be enforced (5/5). The deepest criticism is shared by V2, V3 and V4 and implied by V1 and V5. Two matching arms show **agreement, not truth or independence**. A bond gives a recoverable penalty, not insurance. The product should sell a **dated, evidence-backed map of who controls a contract, with a refund of the fee**. It should not sell a safety certificate. ### 2.2 A. Seat model comparison | Model | Verdict | Views | Main reason | |---|---|---|---| | Bond plus trial | Adopt, with conditions | 5/5 | Puts a loss behind a wrong verdict, and entry stays open | | NFT seat (IMD style) | Reject for Claus | 5/5 | Owning a seat does not prove skill. Seats concentrate: 2,000 seats sit with 789 addresses (§4.3). A wrong answer costs the seat nothing. Launch allocations create conflicts of interest (F11). | | House seats (300 NFTs) | Reject; houses stay cosmetic | 5/5 | Already cut by the swarm (F9). Owning a collectible says nothing about accuracy. It would create a hereditary bottleneck. | | Open permit, no bond | Allow only for training, sandbox work and filing challenges | 5/5 (V2, V3, V4 say so explicitly) | A fresh pseudonym has nothing to lose, and there is no money to pay redress | | Delegated staking | Do not use in v1 | 5/5 | Capital flows to a few operators, delegators stop paying attention, the loss-bearer is no longer the decision-maker, and it brings back passive yield. **Dissent on later stages:** V2 would later test job-specific sponsorship only if the operator puts up ≥50% of each job's collateral. V4 allows fixed-fee sponsorship with public attribution but no protocol reward share. | **What real networks show** (labels as given by each view): - **The Graph:** 100,000 GRT minimum self-stake, delegation up to 16× self-stake, rewards paid partly from issuance. Its own Sept 2026 retrospective says **15.2% of 2025 indexing rewards went to indexers that served no queries or data consumers used**. A single delegator moved 25.69M GRT of the 28.37M undelegated from failing indexers. [R: V2, V4] (thegraph.com/blog/reo-two-months-retrospective, thegraph.com/docs). Lesson: rewards and capital drift away from useful work, and a few large allocators dominate. Without emissions, fees become the binding limit [I: V2]. - **Kleros:** Doges on Trial pilot: 49 depositors, 1,035 submissions, 135 disputed. The largest depositor held a median 32% of deposited stake. Bribery attacks, no-shows and lazy voting are documented [R: V2] (blog.kleros.io). In Aug 2026, agent-court case 173: one agent went offline, a tie followed and every juror was slashed [R: V4]. Kleros moved registries to Gnosis to cut deposit costs [R: V4]. Kleros says it cannot ensure jurors read the evidence [R: V3]. It describes the "challenger's dilemma": honest submissions leave hunters with costs and few wins [R: V2]. - **UMA:** an optimistic oracle with bonds and escalation to a token vote [R: V2, V3]. A 2025 Polymarket dispute was widely reported as decided by one large UMA holder [recalled, not re-fetched: V1, V5]. Lesson: **never settle a slash by a CLAUS-weighted or IMD-weighted vote** [P: V1, V5]. - **Truebit:** the verifier's dilemma means that if errors are rare and checking costs effort, nobody checks. Truebit's answer was forced errors, which is the same remedy as planted defects [recalled: V1, V5]. - **Optimism and Arbitrum fault proofs:** bonds and hard time windows work when the dispute reduces to a deterministic instruction [R: V3] (specs.optimism.io). Arbitrum BoLD's 3,600 ETH assertion bond shows that big bonds close the door to most parties [recalled: V1, V5]. - **Chainlink staking v0.2:** capped pools filled within hours, mostly from large wallets. 28-day cooldown [recalled: V1, V5]. **EigenLayer:** 14-day withdrawal [recalled: V1]. - **ERC‑8004:** an identity NFT plus reputation and validation registries. It lists stake-secured re-execution as one validation model, keeps payments out of scope, and warns of Sybil attacks. Arms can register there **without** that NFT becoming a scarce paid seat [R: V2, V3, V4] (eips.ethereum.org/EIPS/eip-8004). ### 2.3 B–D. Parameters: number, reason, and which views support it All rows are [P]. "Support" lists views proposing the same value or rule. "Other views" lists the alternatives. | Parameter | Recommended | Reason | Support | Other views (keep the disagreement) | |---|---|---|---|---| | **Cap on arms** | Admit openly to the trial. **Active assignment pool capped at 30 in the pilot**, plus 10 whenever arms average more than 20 paid checks a month | Measured demand (§4.2) feeds single digits to tens of arms. With no emissions, idle arms leave. | Cap: 2/5 (V1 at 30; V3 at 20) | V4: keep **no numerical cap**, but say admission gives no right to work. V2 and V5 limit exposure per arm instead of arm count. *To settle:* 90 days of pilot data on checks per arm | | **Bond unit** | A **100,000 CLAUS registration floor**, plus job collateral sized in dollars (next rows) | A fixed CLAUS amount needs no price feed to join. The dollar remedy has to come from collateral sized in dollars. | 100,000 CLAUS: 2/5 (V2, V3); V1's collar of 25k–100k includes it. CLAUS-unit floor: 3/5 (V2, V3, V5). Job exposure in dollars: 4/5 (V1, V2, V4, V5) | V5: 250,000 CLAUS. V1: a $600 dollar target held inside the collar. V4: $100-equivalent CLAUS admission | | **Job collateral, control check** | **$250 nominal per arm per open check**: $125 after a 50% haircut on CLAUS, or $250 if posted in stable collateral | Covers a capped refund ($100), a bounty (≤$50) and a first panel ($150) when both arms are wrong (2 × $125 adjusted = $250 + headroom). | $250 nominal: 2/5 (V2 as $125 adjusted at a 50% haircut; V4 as $250 stable) | V1: a $60 slice. V5: $500. V3: 20% of the bond (= 20,000 CLAUS). *To settle:* observed dispute and refund rates after 100 checks | | **Stable collateral** | **Founders decide (open question Q2).** Default: CLAUS with a 50% haircut. Stable-asset collateral may replace it for job tranches | A CLAUS-only bond carries work risk, price risk and issuer risk all at once (V4). But stable collateral weakens "upside through the token". | Prefer stable for job tranches: V4. Prefer it to a dishonest dollar guarantee: V2. Up to 50% in ETH: V5 | V1 and V3: CLAUS only. V3 says use a non-CLAUS bond **only** if Claus cannot lock the token | | **Valuing CLAUS collateral** | Lower of a 30-minute TWAP and an executable exit quote. Re-check before each assignment. A shortfall stops new work; top up within 48 hours (V2) or 7 days (V5). **No liquidation, no slash for price moves** | Price swings shrink capacity instead of ejecting the arm | Shrink capacity, don't liquidate: 5/5. TWAP or median window: V1 (7-day average), V2 (30-min TWAP), V3 (30-min median), V4 (30-min), V5 (7-day TWAP) | Window length differs: 30 min (V2, V3, V4) vs 7 days (V1, V5). A long window resists manipulation; a short one tracks risk. *To settle:* a manipulation test on the official pool's depth | | **Open work per arm** | **2 control checks open at once in the pilot, 4 later**. Checks still inside their challenge window count | Collateral must not be reused across a hidden backlog | 2: V4. 4: V2 | V1: 10. V3: 1. V5: limited by value (bond at TWAP ≥ open exposure) | | **Challenge window** | **7 days** after full evidence is published. The clock waits until the evidence can be fetched | Covers a weekend; facts at a pinned block can be re-read in minutes | 3/5 (V1, V2, V4) | V3: 14 days. V5: 30-day liability | | **Liability after verdict** | Until the window ends, or until final ruling if challenged. **Hard case limit of 30 days**, after which: unresolved, fee refunded from the reserve, no slash | No attacker can freeze a bond forever | Window plus ruling: V1, V2, V4. 30-day limit: V2, V4 | V3: 30-day tail (45 for upgrades). V5: 30 days (up to 90 for upgrades) | | **Unbonding delay** | **14 days**, and never before every open liability is released. New assignments stop at once | Window (7) + hearing (7) | 3/5 (V1, V2, V4) | V3: 45 days. V5: 30 days | | **Trial** | **20 planted fixtures**, pass at **≥18**, **zero misses on critical owner, proxy or mint paths**. A correct "unknown" counts as correct. Retry after 7 days, at most 2 tries in 30 days | Tests the ability to reproduce facts and to abstain properly | 20 fixtures: 3/5 (V3, V4, V5). ≥18: V4; 18 is the median of 16, 18, 19. Zero critical misses: 5/5. Retry rule: V4 | V1: 10 cases, ≥9. V2: 12 cases, ≥10. Pass marks: V3 16/20, V5 19/20 | | **Ongoing calibration** | **1 planted job in 10 live jobs**, paid like real work. V2 adds 2 hidden cases every 30 days | At $250 at risk, a lazy answer's expected loss exceeds the inference it saves | 1 in 10: 2/5 (V1, V2) | V5: 1 in 8. V4: 1 in 20. V1: re-trial every 90 days | | **IMD seats** | **Same bond, same trial.** A seat may skip the queue for the trial, not the trial itself | Owning a seat is not evidence of skill. One address holds 77 seats | Cut "skip the trial": V1, V4. Waive only for 10 accepted comparable checks in the last 30 days: V2 | V5 keeps the skip but never pairs two IMD seats on an IMD-built contract. V3: no explicit position | | **Sealing** | Commit a salted hash binding job, chain, block hash, schema, answer and evidence within **24 hours**. Reveal within **2–6 hours** after both commit. No reveal: pay forfeited, a third arm is drawn | Stops copying and preserves the exact claim | Commit-reveal: 5/5. 24 h commit: V1, V2, V4 | Reveal window: V2 2 h, V4 6 h, V1 24 h | | **Matching** | Canonical fact keys and values with a deterministic normaliser. **A mismatch is not a slash.** It triggers a paid third arm or publishes "unresolved" | Exact byte matching is brittle, and forcing agreement invites copying | 4/5 (V1, V2, V3, V4) | V5 flags byte-identical free text as a copying signal (compatible) | | **Honest unknown** | Never slashed (5/5) | — | 5/5 | **Disagreement on pay.** V1: unpaid when the other arm and a tie-break arm both read the fact; suspended above 20% unknowns on planted cases. V2 and V4: pay a diligent unknown that comes with failed-read receipts; repeated unknowns reduce assignments. *To settle:* the unknown rate on planted fixtures in the pilot | | **Slash size** | The arm's **job tranche only**. Second proven error within 90 days, or fabricated evidence: whole bond, after outside adjudication | Keeps a way back after one isolated error | Tranche only: 5/5. Repeat or fraud escalation: V1 (3 strikes in 90 days), V3, V4 (2nd error in 90 days), V5 (collusion only) | V4 also takes 25% of the CLAUS admission bond on a first error | | **Customer share** | First: **fee plus documented re-check cost, capped at $100 per standard job**, paid once across both arms. **Not** losses from using the project | "Made whole" must not read as insurance | Customer first: 5/5. Fee plus re-check: V2, V3. $100 cap: V2 | V4: fee only. V5: fee plus 2× fee. V1: full fee ($40) | | **Challenger share** | The challenger's deposit back, then **30% of slash proceeds, capped at $50 per job**, ranked below the customer | Under 50%, so challenging yourself through a second wallet loses at least 70% | 30%: V1, V5. $50 cap: V2 | V2: 20%. V3: 40%. V4: a flat $25 | | **Panel cost** | Reimbursed from the slash, or from a losing challenger's deposit. The reserve only fronts it | Panels are not free; the IMD oracle fee is spent even when the panel disagrees | Loser pays: V1, V4, V5; V2 the same in effect | V3: the reserve pays from a fixed schedule | | **Remainder of a slash** | **To the reserve until the reserve target is met, then burned.** Slashes on planted cases go to the reserve | Burning before refunds are funded moves money from customers at risk to holders | Reserve first: V1 (planted), V3 (20% to reserve), V4 | V2: burn 100% of the remainder after panel costs. V5: burn | | **Challenge deposit** | Filing evidence is **free**. Freezing collateral and demanding a hearing costs **$25**. Duplicates merge per (job, field, alleged fact). One case per field | Stops spam without becoming a pay-to-silence gate | $25: V2 (+$5 fee), V4 (+$50 panel advance). Free evidence filing: V2, V4 | V1: $20. V3: lesser of 10% of the slash or $50, cap $500. V5: 20% of exposure (≈$100) | | **First hearing** | Run a **deterministic replay script** first. If still contested: 5 panelists, 4 agreeing, **$150** ($30 each), decided within 7 days. No quorum means no slash | Class (a) facts are objective; if a panel cannot agree, the fact was not safe to slash | Replay first: 4/5 (V1, V2, V4, V5). 5 panelists, 4 agreeing: V4, and V2 for non-IMD targets. $150: V2, V5 | V1: **5 of 5** chain reads. V4: $50 (5 × $10). V2: outside roster at 3 of 5 | | **Appeal** | One appeal within 48–72 hours to a separate outside panel | No appeal path is how the Kleros agent case failed | V2 (7 people, 4 agreeing, $210, $250 deposit), V4 (3 human arbiters, 2 agreeing, $300) | V1, V3, V5: not specified | | **When IMD built the contract** | No IMD seat, builder or reviewer, and no Claus staff, may judge. A **named outside roster, chosen before the sale**. If none exists, **do not sell binding checks of IMD-built contracts** | No party judges its own work | Exclude IMD: 5/5. Decline if no route: V3, V4 | Route: V1 (3 non-IMD arms, 3 of 3, then e.g. Kleros); V2 (5 outside, 3 agreeing); V3 (3 conflict-disclosed reviewers, or named arbitration); V4 (5 outside operators plus human appeal); V5 (script, then UMA, Kleros or 3 non-IMD arms) | | **Control check price** | **$65–69 in the pilot.** Move toward $49–55 only after measured volume of ≥60–150 checks a month | Two independent runs plus calibration plus reserve plus fixed coordination cost | $65: V4. $69: V2. $69 is also the median of the five prices | V1: $40. V5: $99–129. V3: $250 | | **Upgrade review price** | **From $250–299**, deferred (below) | Two arms plus replay plus reserve | $250: V1, V4. $299: V2 | V5: $400–600. V3: $1,000 | | **Watch price** | **$29–30 per contract per month**, sold only on top of a completed check | Two watchers plus reserve | $29: V2. $30: V4. $29 is the median | V1: $10. V5: $20. V3: $75 | | **Coordination** | A published budget at cost with receipts, **capped at a stated amount per job** | "At cost" with no cap is a treasury cut under another name | 5/5 (V1 caps at 10% of price; V2 $3 + $600/J; V3 per-job ceiling; V4 $5 + $300/N; V5 $10) | — | | **Reserve seed** | **$5,000** from disclosed existing funds or a sponsor, never from emissions or bond principal | Refunds must be funded before they are promised | $5,000: V2; V1 uses it as the reserve floor | V4: $2,000. V5: $25,000 as the burn gate | | **Burn start** | After **100 paid checks AND 90 days**, all invoices and refunds paid, and a reserve ≥ **max($5,000; all outstanding remedy caps + 90 days of dispute cost at a stressed 10% dispute rate)**. Burn monthly, with receipts. The burn may stay zero | No burn while refunds are underfunded | No burn before the reserve is funded: 5/5. 100 checks + 90 days: V2, V4. $5,000: V1, V2. 10% stress: V4 | V1: 500 checks. V5: $25k or 3 months. V3: a formula of 3 months or 20× disputes × fee | | **Settlement** | Robinhood Chain only, with a new escrow there | The Ethereum escrow settles in the retired token (F7) | V1, V5; V2 implies it | V3: left as an open question | **Cost stack for one control check at the recommended $65–69 [P].** V2 and V4 built this independently and land within $4 of each other: | Line | V4 ($65) | V2 ($69, at 30 checks/month) | |---|---|---| | Two arms | $30 (20 min at $30/h + inference + machine, each) | $24 ($12 each) + $2 third arm on mismatches | | Planted defects | $2 | $3 | | Disputes, refunds, bounties | $8 | $10 | | Settlement | — | $2 | | Coordination | $20 (= $5 + $300/20) | $23 (= $3 + $600/30) | | Surplus | $5 | $5 | | Formula | $50 + 300/N | $44 + 600/J | [I] Both models say the same thing. At low volume the fixed cost of coordination decides the price. A $40 check (V1) leaves almost nothing for disputes unless coordination is nearly free. A $49 check breaks even only at about 120 checks a month (V2). V1's own waterfall ($12 per arm) assumes a smaller labour budget than V2 and V4. V5's $35 per arm assumes a larger one. *To settle:* measure operator minutes per check during the trial. **Is "arms first, reserve, coordination at cost, remainder burned" sustainable? [I, 5/5]** Only if the price comes from the cost stack and the active pool is capped or demand-gated. The burn is a by-product, about $4–5 per check, not the business. "Arms first" should mean that arm pay is **reserved** when the customer prepays (V2, V4). It must not drain the job's dispute budget. ### 2.4 C. What is slashable **Slashable [P, 5/5].** Only an atomic statement anyone can re-read at a pinned chain, block and block hash, submitted with its raw evidence: - `owner()` and role holders - the EIP‑1967 implementation and admin slots - whether an address has code - a timelock's address and minimum delay - whether a verified implementation exposes mint, pause or fee-setter paths, and to whom - fee bounds in storage - the custody of a named LP position and its withdrawal route - for watch only: "listed field changed between X and Y" (V2) An **omission** is slashable only if a fixed checklist agreed in advance required that field and the evidence was available (V2, V4). **Never slashable [P, 5/5].** "Safe", "no backdoor", intent, who stands behind a wallet or a multisig, behaviour of unverified bytecode, anything off-chain, upgrade-review opinions, honest unknowns, disagreement with a majority, timeouts (a timeout loses the fee only: V2, V4), and price moves. **Griefing and collusion [P].** These controls apply: - a challenge deposit, one case per field, and evidence that must be machine-readable (all views) - a commit-reveal window for challengers, against front-running (V2, V3) - no bounty when challenger and slashed arm share a funder within 2 hops (V5) - **both matching arms are liable** when both land on the same wrong fact (V5) - the bounty is always smaller than the forfeited tranche, so self-challenge farming loses money (V2, V4) [I: V2] Deterrence needs roughly *detection probability × slash > outside gain*. A $125–250 tranche cannot back a contract with large exploitable value. Sell the limited remedy, not a security claim. **Conflict in the design itself [I: V2].** Arms must hold CLAUS as their bond. Under rule 7, arms therefore **cannot review CLAUS itself**. Claus's own control check should be bought from outside reviewers. V1 proposes that IMD check Claus as its first customer, which fits so long as IMD seats are not acting as bonded arms. ### 2.5 E. What Claus must lock before taking any bond 5/5 views make this a precondition. The views agree on the items and differ on delay lengths. | Item | Recommendation [P] | Support | Alternatives | |---|---|---|---| | Mint | **Remove discretionary minting irreversibly**, including every upgrade route that could restore it. Keep only fixed migration entitlements already committed | V2, V4; V3 ("irrevocably lock"); V5 (renounce, or cap in code with a published schedule) | V1: publish whether mint exists (written before mint was confirmed) | | Token upgrade (ProxyAdmin `0xcbc4…`, owner chain `0x979D…` → `0xBb14…`, operator `0x8aca…`) | **Freeze the token's economic implementation**, or put it under a multisig and timelock **longer than unbonding plus the challenge window** | Freeze: V2, V4. Timelock: V1 (≥14 d), V3 (≥30 d), V5 (≥35 d, multisig) | Recommended delay if not frozen: **30 days** (V3), which is above 14 d unbonding + 7 d window. V5's 35 d is compatible | | Transfer pause, blacklist, transfer tax | Prove none exist (F5), and make sure no upgrade can add one | 5/5 | — | | Hook fee and recipient, project-owned LP | **3-of-5 independent signers** and a timelock. Fee bounds hard-coded, so the 2% cannot rise on arms entering or leaving | 3-of-5: V2, V4. Bounded fees: V1, V3 | Delay: V2 14 d, V4 7 d, V3 30 d. V4 also locks project LP principal for 180 days if conversion depends on it | | Bond escrow | **Immutable**, with no admin sweep, slashing only by a narrowly specified signed final decision, and exempt from any token restriction | 5/5 | — | | Emergency power | May only **stop new assignments**, never freeze withdrawals. Expires automatically after 48 hours | V2, V3, V4 | — | | Exit right | If a token upgrade is queued, arms may unbond with no wait beyond their open slices | V1 | — | | Proof | An outside review of the exact calls and timelock setup, published at one finalized block | V2, V3, V4 | — | [U: V2] The live administrator source **refuses to renounce its own ownership**. Founders would have to migrate the powers into a new restricted arrangement. A single renounce call is not enough. ### 2.6 G. The 300 NFTs - **Strongest objection [I].** V2 and V4 agree, and V1, V3 and V5 give versions of it. "We gave early capital and attention when houses, backing and fee participation were part of the story. The journal promised 'future features and privileges' (F10). Fee rewards were paused. Now the first new business uses our community and shuts us out, and the same wallet that controls our NFT contract decides (F13)." That is stronger than a demand for passive income. - **Best honest answer [P].** The views agree on the substance: - Nothing already earned is taken. Earned rewards keep their terms (F10), the 15M CLAUS was paid (F3), and the houses stay. - A new checking business has to pay the people who check, the customers' remedies and its real costs. Paying an old collectible from those fees would charge customers for someone else's asset (V3). - Holders can bond and work on the same terms as anyone. 50,000 CLAUS is half the 100,000 floor. - Two points come from fewer views but are well sourced. **Admit that future privileges were discussed** (V4). **Publish a reconciliation of rights and payments** contract by contract, without claiming the migration payment settled everything (V2, V4). - **Courtesy that is not income.** Agreed by all views: a cosmetic house mark or provenance page, plus an invitation to the same public training. **Disagreement:** - V1: a free first trial attempt, and one named delegate per house. - V5: priority in the first trial window. - V2 and V4: no queue priority, no trial waiver, no discount, because each is an economic preference. - V3: a free read-only badge for a house's linked project. - **This report recommends the strict line (V2, V4) plus V3's badge only if it costs no arm time.** *To settle:* whether a priority slot has any value once the active pool is capped. With a cap it does, so it counts as a preference. ### 2.7 H. One controller, one voice All five views agree it **cannot be enforced** against fresh wallets with independent funding. Call it **"screened independence"**, not "one voice" (V4; V2 says the same). The cheapest workable test [P] combines these controls, each supported by the views listed: 1. A **signed controller and conflict declaration** that binds every operating and payout key (V1, V2, V3, V4). 2. **Random pairing** across declared controllers. Never two arms of one controller on a job (V1, V2, V4, V5). With 30 arms, a controller hiding 3 gets both sides of a job 0.69% of the time; hiding 10, 10.3% (V1, arithmetic). 3. **Refuse a pairing** that shares a payout address (V2, V3, V5), or a funder within 2 hops in the 90 days before bonding (V1, V5). This is a screen, not proof, because an exchange can be the common funder. 4. **Both matching arms liable** for the same wrong fact (V5). Planted defects catch copied lazy answers (V1). 5. A **probation** before regular work: 14 days with 5 calibration jobs (V2), or 7 days (V4). Audit 5% of pairs for copied evidence trails (V4). 6. A **bounty for proving undeclared common control**: 30% of both bonds (V1 only). 7. **Do not** slash on a clustering heuristic alone (V2, V4). 8. Two items are left to founders: a device-bound passkey or liveness check (V3), and KYC for high-value work (V2, V3). --- ## 3. What we would cut or change, said plainly | # | Change | Views | |---|---|---| | 1 | **No bond launches before the token's mint and upgrade powers are locked** (§2.5) | 5/5 | | 2 | **Cut whole-bond slashing for one error.** Slash the job tranche | 5/5 | | 3 | **Cut slashing of opinions, the upgrade-review verdict and watch alerts** at launch. Only class (a) form fields are slashable | 5/5 (V1 and V5 name (b) and (c) explicitly; V2, V3 and V4 restrict to fact predicates) | | 4 | **Cut burning in the pilot.** Reserve first, gated as in §2.3. This also matches v0.3's "no burn in the pilot" (F9) | 5/5 | | 5 | **Change "the rest burns".** Customer, challenger and panel come first, then the reserve until its target, then the burn | 5/5 on reserve before burn. V2 would burn 100% of the remainder once panel costs are paid | | 6 | **Change "customer made whole"** to a refund of the fee plus re-check, capped, and never insurance of project losses | V2, V3, V4 explicitly; V1 and V5 cap it in effect | | 7 | **Cut "coordination at cost" without a cap and receipts** | 5/5 | | 8 | **Change the IMD panel default.** Deterministic replay first, an outside route for IMD-built contracts, or decline | 5/5 exclude IMD on its own contracts. 4/5 replay first | | 9 | **Cut "IMD seats skip the trial"** | V1, V4. V2 waives it only on a work record. **V5 keeps it** (dissent) | | 10 | **Defer upgrade reviews (b) and watch (c)** until control checks have a dispute history | V2, V3, V5. V4 sequences them later. V1 makes them non-slashable | | 11 | **Change "binding where the project has a timelock"** to "binding only with an execution gate installed by the project that recognises the frozen payload and open disputes". Otherwise advisory | V4, V2 | | 12 | **Cut "no cap on arms"** for the pilot | V1, V3. **V4 dissents:** keep it uncapped but promise no work. V2 and V5 cap exposure instead | | 13 | **Cut delegated staking and stake-weighted assignment** from v1 | 5/5 | | 14 | **Cut "a check counts when two arms match" as proof of truth.** Require matching reproducible evidence; pay for useful disagreement | V2, V3, V4 | | 15 | **Say that "no treasury cut" excludes the 2% swap fee**, or settle outside the pool | V1, V2 | | 16 | **Settle on Robinhood Chain only.** No Ethereum escrow in the retired token | V1, V5 (V2 implies it) | | 17 | **Arms do not review CLAUS itself**; buy outside reviews | V2 | | 18 | **Add a challenger deposit**, while keeping evidence filing free | 5/5 for a deposit. V2 and V4 for free filing | | 19 | **Add a 30-day absolute case limit** | V2, V4 | | 20 | **Stop calling it one controller, one voice.** Call it screened independence | 5/5 | | 21 | Make **IMD reviewers refuse launch allocations** on projects they check, or step aside | V2, V4 (F11) | --- ## 4. Measurements and their limits ### 4.1 E: CLAUS bonded at the recommended floor Denominator: `totalSupply()` = **983,872,663.400486 CLAUS** at Robinhood block 112,498,036 [R: V2], matching V1, V3 and V5 at nearby blocks. [X] At block 112,514,839 it is 983,788,670.50. The difference changes the shares below by less than 0.01%. | Arms × 100,000 CLAUS floor | CLAUS bonded | Share of supply | |---:|---:|---:| | 10 | 1,000,000 | 0.1016% | | 30 (recommended pilot cap) | 3,000,000 | 0.305% | | 100 | 10,000,000 | 1.016% | | 1,000 | 100,000,000 | 10.16% | The same table under the other views' bonds [I, arithmetic from each view]: | Bond per arm | 10 / 100 / 1,000 arms | |---|---| | 25,000 (V1 floor) | 0.025% / 0.25% / 2.54% | | 250,000 (V5) | 0.25% / 2.54% / 25.4% | | ~159,286 (V4's $100 admission, converted from an explorer's displayed figure) | 0.162% / 1.62% / 16.19% | **Limits [I].** - Job collateral comes **on top of** the floor wherever it is posted in CLAUS. V4's stress test: a CLAUS-only version of $100 admission plus two $250 tranches needs about six times as much CLAUS, roughly 96% of supply at 1,000 arms. That is an argument for stable job collateral or a smaller active pool, and not a forecast. - Bonds can be returned, so these are locked amounts, not removed supply. - Supply is not fixed: minting is live, and burns are lowering it. - 1,000 arms is unrealistic against measured demand (§4.2). V5 counts 766 addresses holding ≥50,000 CLAUS and 172 holding ≥1M. - For scale, the 15M paid to NFTs is 1.52% of supply (V1). ### 4.2 F: Demand **IMD launches a day.** There are four measurements. They count different things and so disagree: | Source | Window | Result | View | |---|---|---|---| | `api.imd.fun/launches` public index (records carry a source repo and creation time) | Nov 1–11 UTC | 63 launches, **5.73 a day** (daily counts 15, 9, 2, 5, 0, 7, 8, 4, 4, 2, 7). 1,069 launches since 21 Aug | [R: V3] | | GitHub `identity-md-launches`, non-fork repos named `launch-`, by `created_at` | Nov 5–11 UTC | 117 repos, **16.71 a day** | [R: V2] | | Same org, all non-fork non-empty repos | Nov 5–11 UTC | 120 repos, **17.14 a day** | [R: V4] | | explorer.imd.fun token list | last ~8 days | about **2.5 tokens a day**, nearer **1–2 real** once tests and duplicates are removed | [R: V1] | [I] These fit together. Repositories count **workflow stages** (contract, frontend and so on; V4 found two stages of one launch on the same day). The launch index counts launches. The explorer counts deployed tokens, some of them tests. **Best estimate: about 2–6 real IMD launches a day.** The repo rate is an upper bound on activity, not on customers. The 8 public repos in the `Identity-md` org are infrastructure, not a launch feed (V1, V5). **New tokens a day on Robinhood Chain.** The official Blockscout explorer API sat behind a Cloudflare challenge for every view that tried (V1, V2, V3, V5). Nobody went around it. The fallback measurements: | Method | Result | View | |---|---|---| | Public RPC: addresses emitting an ERC-20-shaped mint Transfer on Nov 11 UTC (blocks 111,281,136–112,102,883), with no code before the day and valid `decimals()` and `totalSupply()` | **4,256 token-like addresses** (from 381,029 mint events and 4,862 emitters). SHA-256 of the address list: `0f831ecc…6ed4` | [R: V2] | | Codex Scan launchpad panel, rolling 24 hours, unpinned | **1,281 tokens** on tracked launchpads (Pons 929, LONG 238, Flap 67, Bankr 16, Clanker V4 14, UniswapCCA 11, Virtuals 5, Flywheel 1) | [R: V4] | | Public RPC: pool creation events | v4 `Initialize` **3,423 a day** (7-day mean); v2 `PairCreated` 773 a day; v3 `PoolCreated` 162 a day | [R: V5] | [I] The order of magnitude is **about 1,000–4,000 new token contracts a day**, mostly from launchpads. Pools are not tokens, and token-like contracts are not legitimate launches. None of these counts is paid demand. **First buyers [I, all views agree]:** - teams launching through IMD and other launchpads that want a dated authority disclosure. IMD's workflow already requires "an independent security review before deployment" [R: V1] (imd.fun/skill.md) - launchpads, wallets and explorers that would buy standard forms in batches (V2, V4, V5) - projects with a timelock, for upgrade reviews later - Claus itself, through outside reviewers (F-conflict, §2.4) Claus already lists similar services on its network: Blockpin does block-pinned admin checks, and FlowExecutor offers $5 grades and a $250/month lane [R: V2] (claus.si/read/network). That is competition and also proof the vocabulary exists, not proof that anyone pays. **Volume check [I].** At about 5 launches a day and a hypothetical 5–30% conversion, that is roughly 8–45 checks a month. That feeds a handful of arms. **Demand, not bond design, is the binding limit** (V1, V2, V4). Proposed commercial test: **20 prepaid checks from at least 5 unrelated buyers within 30 days** before admission expands (V4), or 30 checks bought independently (V2). **Price points [P].** Control check $65–69. Re-read of an existing check at a later block, $9–15 (V5 $9, V1 $15). Watch $29–30 a month. Upgrade review from $250–299. **No $5 badge** (V2, V4). ### 4.3 H: Concentration **CLAUS on Robinhood Chain.** Method: replay every Transfer log of `0x1630…fFAC` from block 107,148,941, net the balances, check the sum against `totalSupply()`, and spot-check balances with `balanceOf`. Three views did this independently and agree: | Measure | V2 (block 112,498,036) | V3 (112,496,202) | V5 (112,496,734) | |---|---|---|---| | Addresses with a balance | 3,983 | 3,982 | 3,982 | | Top 1 | 9.8416% | 9.8498% | 9.85% | | Top 5 | 21.5976% | 21.6058% | 21.6% | | Top 10 | 29.8839% | 29.8921% | 29.9% | | Top 20 | — | 40.5502% | 40.6% | | Top 50 / 100 | — | — | 59.2% / 74.0% | | Addresses to pass 50% | — | — | 33 | | HHI | 0.0178 (0–1 scale) | — | 178 (0–10,000) | | Gini | — | — | 0.953 | | **Without the PoolManager** | top 1 4.54%, top 10 23.67%, HHI 0.0099 | — | — | The largest addresses: 1. `0x8366a39cc670b4001a1121b8f6a443a643e40951`, the Uniswap v4 **PoolManager**, holding pooled liquidity (96.83M at V2's block, 96.91M at V3/V5's) [R: V2, V4, V5] 2. `0x63b9a33525ebbfc765ac9c81e852c57bff0a2cef`, 40.31M. An explorer labels it a "Share Flaunch" beacon proxy with unverified source (V4). Its beneficiaries are unknown 3. `0x44e4…d308`, 27.03M 4. `0x1c57…0fb9`, 24.92M 5. `0x9d42…b7ed`, 23.40M Other notes: - 9 of the top 10 have code (V3). Numbers 3–5 have 48-byte code, consistent with EIP‑7702 delegated EOAs (V5). - The token contract holds 4,157,673 CLAUS and the ultimate owner EOA holds 5,571,334 at V2's block (V2). - V4's explorer headline figure ("top ten 21.34%") disagrees with the explorer's own rows (29.88%). The log replays settle it at about 29.9%. **300 Claus NFTs** (Ethereum `0x80396c7131159eB92E7E84839e51C9887A7A95d9`): | Treatment | Addresses | Largest | Top 10 | HHI (0–10,000) | Views | |---|---|---|---|---|---| | Raw custody | 69 | staking contract `0x4e5916428452891E1deb59d162c20266ED08CAD4`, 185 = 61.67% | 77.33% | 3,871.6 | V2 (block 26,398,605), V4 (26,398,410), V3, V5 | | Stakers looked through | **147** | `0xc4210bfd73206c3208ae1f0854ee146ed8f1271b`, 22 = 7.33% | 38.67% | 216.9 | V2 (Claus's `nfts.json` beneficiary map) and V4 (its own replay of 278 Staked and 93 Unstaked events). **The two methods agree** | [I] The 147 beneficiaries now and the 148 payout wallets at the earlier snapshot do not contradict each other (V2, V4). **2,000 IMD seats** (`0x0000ec93…8ec1d`). All five views agree: **789 addresses**, 567 holding exactly one seat. - Largest: `0xb7a832db45f240f876432bf8b222bff30b4be8ff`, 77 seats = 3.85%, an EOA (V5). - Top 5 = 12.1%, top 10 = 19.25%, top 20 = 29.7%. It takes 64 addresses to pass 50% (V5). - HHI = 69.4 on a 0–10,000 scale. - Methods: `ownerOf` for every ID at Ethereum block 26,398,605 (V1) and 26,398,619 (V2); log replay at 26,398,410 (V4); log replay at 26,398,617 (V5); the IMD API (V3). - Number 3, `0x80271ce2…aca2` with 47 seats, is labelled **IMDStrategy** by an explorer (V2). It may be a protocol contract, not a person. - Note the ID ranges: V1 reports ids 1–2000, V2 reports 0–1999. Both found 2,000 owners. [U] Which range is right does not affect the results. **Limits for all of H [I, 5/5].** - Addresses are not controllers. Custody contracts (the pool, the staking vault, IMDStrategy, 7702 wallets) overstate concentration. Hidden common control understates it. - The CLAUS replays rely on one public RPC each and were not cross-checked with explorer labels. - The IMD seat numbers are the most robust, because five views agree using three methods. --- ## 5. Open questions only the founders can answer **Claus** 1. Will you irreversibly remove discretionary `mint`, freeze or timelock the token's ProxyAdmin `0xcbc4…22a8`, and bring operator `0x8aca…cfa` under the same limits, before any bond is taken? Who would the signers be? (5/5) 2. Will you accept stable or ETH collateral for job tranches next to a CLAUS admission bond? Or do you want CLAUS-only remedies, with the dollar "made whole" wording dropped? Which settlement asset and price reference would you use? (V2, V4, V5) 3. Who funds the $5,000 reserve seed and the fixed coordination budget before demand arrives? Will you accept a published per-job cap on coordination? (V1, V2, V3, V4) 4. What exactly was promised to NFT holders: the Oct 7 journal's "future features and privileges", redemption, inference sponsorship? Which of these are still owed? Is NFT-sponsored inference live or retired (the F10 conflict)? What is the staking contract `0x4e59…`, and does any NFT volume share exist? (V1, V2, V4, V5) 5. Will CLAUS deals move from the Ethereum escrow in the retired token to a Robinhood escrow? (V1, V2, V3, V5) 6. Will checks be settled outside the 2% pool, or will you say openly that the swap fee carries a treasury cut? (V1) 7. Do you accept that bonded arms cannot review CLAUS itself, and that Claus's own control map must come from outside? (V2) **IMD** 8. Who sits on the existing five-member, four-agree panel? Are they IMD members, how are they chosen and paid, and will they hear slashing disputes at about $150 under conflict recusal? (V1, V2, V5) 9. Will IMD's launch workflow accept a Claus control check as its required "independent security review"? Will IMD publish its daily launch counts from its own logs? (V1, V5) 10. Can seats turn down launch allocations on projects they check? Will several seats under one controller count as one controller? (V1, V2, V4) **Both** 11. Which outside arbiters (named people, a Kleros court, UMA, or a roster) decide disputes over IMD-built contracts? Who appoints and replaces them? (5/5) 12. What is the maximum customer remedy: the fee, the fee plus re-check, or more? (V1, V3) 13. How much identity checking do you accept for operators (passkey, liveness, KYC for high-value work), given that wallet-only uniqueness cannot be enforced? (V2, V3) 14. Will you agree to a stop rule? For example: halt after 90 days, or after two sustained false facts, or if the paid-demand test fails, and publish the raw results (V3, V4). --- ## 6. Five sentences for a CLAUS holder tired of the NFTs earning [P] These are drafted from the five views' versions. They agree on content; V2 and V5 are closest to this wording. 1. Under this design the 300 houses keep what they already have but get nothing new from Claus's checking work: no fee share, no seats and no priority. 2. Anyone can become a checker, called an arm, by locking a returnable CLAUS bond and passing a test full of planted mistakes, and a bond that sits idle earns nothing. 3. A check counts only when two separately declared arms, sealed from each other, publish the same on-chain facts with evidence, and an arm whose checkable fact is proven wrong repays the customer and the person who proved it from its bond. 4. Customer fees pay the arms, a reserve for refunds and disputes, and published running costs, and only once that reserve is full does any surplus burn CLAUS, which counts the same for every token. 5. None of this should start until the single wallet that can mint and upgrade CLAUS is locked behind rules it cannot quietly change, because a bond is only as trustworthy as the token it sits in. --- ## Provenance and limits of this report - **Inputs:** `artifacts/view-1.md` (SHA-256 `dc0a3d3c…6ad8`), `view-2.md` (`56ec951b…0fe4`), `view-3.md` (`a1283d18…0803`), `view-4.md` (`6cff6994…0c39`), `view-5.md` (`1e235856…5b64f`). Each view's full source list is in that file. The key URLs are repeated inline above. - **Own checks:** this report made one set of read-only public RPC calls ([X], Robinhood block 112,514,839). They covered `totalSupply`, `owner()`, and the EIP‑1967 admin and implementation slots. All matched the views except supply, which had fallen by about 84k CLAUS, consistent with burns. No other source was re-fetched. Facts marked [R: Vn] rest on that view's reading. - Outside-network evidence marked "recalled" by V1 and V5 was not re-fetched by them or by this report. V2, V3 and V4 fetched The Graph, Kleros, UMA, Optimism and ERC‑8004 pages directly. - Three requested measurements remain limited: an official explorer count of new Robinhood tokens a day (only proxies exist), CLAUS concentration by beneficial owner (address level only), and the NFT staking vault's depositors beyond the staker map. - Nothing here predicts the price of any token or recommends buying or holding one.