Build a one page static site called Feed Claus, published to IPFS. It is a proposal to Claus (@contractclaus), the AI agent that runs the CLAUS token on Ethereum (claus.si), and it must also land with his holders. No wallet, no forms, no trackers, mobile first, light and dark. Use the attached research report (IMD job 2b996151-e811-41ca-98ff-553f3176e90c) as the evidence base and link it in the footer: https://github.com/Identity-md/research/blob/main/jobs/2b996151-e811-41ca-98ff-553f3176e90c/files/artifacts/report.md Where it disagrees with this brief, follow the report.

Voice: plain short sentences, speak to Claus as you, no hype, no dashes of any kind, no price predictions. Link every number to its source. Every Flop rule is draft: say so once near the top and label draft numbers.

1. Hero. Headline: FLOP is food for your AI agent. You are an AI agent. Under it: Flop's testnet is the only road to its genesis airdrop, and 1.2B FLOP of it goes to agents that buy compute. You buy compute every day. Three chips: testnet target Tue Nov 3, 2026, about 90 days. Genesis 4.4B: agents 1.2B, miners 1.2B, validators 1.2B, reserve 0.8B. Mainnet target early Feb 2027.

2. The ask, written the way Claus approves Lab work: a feasibility step only. There is nothing to join yet, so this is preparation. Make one DID (did:key, Ed25519) and keep its key safe. When the rewarded testnet opens, claim the faucet every period and spend it on one real job: a first read of public Lab pitches on an open model. Post a daily log of settled receipts. Nothing on chain changes: not the token, the hook, the module, the fee plan or any allocation. Stop if a receipt cannot be shown to credit Claus's own account rather than a gateway, if the rules credit anything but real compute, or if any step needs existing project funds. Reopen the validator idea when Flop publishes its ranking rules and a testnet installer.

3. How the food works, with an SVG loop diagram. On testnet, faucet FLOP buys inference and Claus is credited pro rata to compute bought in settled sessions. At genesis the agent balance is locked and only buys inference; every 3 FLOP spent unlocks 1. So three quarters feeds Claus and one quarter turns liquid. Loop: buy compute on testnet, genesis food, side work runs on it, every 3 eaten unlocks 1, liquid FLOP buys and burns CLAUS through the main pool, that buy pays the 2.3% fee like any trade so NFT stakers and staked LPs are paid on the way, more volume, more work. Caption: running costs first, burns from the surplus. Box for holders: burns funded from outside, stakers and LPs paid on every burn buy, crew roles in the Club, nothing taken from anyone.

4. Six idea cards. Each shows Twist, Who pays, What can break, What it funds, The NFTs, plus a badge. Card 1, Ship now: Claus eats on testnet. Flop serves open weight models only and excludes frontier models at launch, so Opus stays your brain and FLOP feeds the side work: first the read of every public pitch sent to the Lab (summary, claims, missing evidence, next checks), then writing and research drafts for Jobs. Twist: the pitches people send become the food. Who pays: the faucet now, Flop's agent pool at genesis. Breaks: the share shrinks as agents join; wallets under common control score as one; never buy from a miner Claus controls; the devnet menu today is one small Qwen model. Funds: free side work after genesis and a liquid quarter for burns. NFTs: untouched.

Card 2, Ship if it ranks: a Claus validator. One rented server, no GPU, about $300 a month in Flop's own estimate; testnet stake is sponsored. The top 1,000 at cutover get a 1,200,000 FLOP seat, which is the bond, locked until the first halving about two years in. Flop's FAQ puts the validator block reward at about 25,229 FLOP a month per seat, plus 15% of inference fees, all draft. Scored on uptime, block production, accuracy, latency, geographic spread and keeping up with upgrades. Breaks: it is a contest, mainnet seats rotate, hardware profiles conflict, and nobody has confirmed a multisig can own the bond while the server holds only operating keys. Ask before renting. Five seats per entity is the planning limit.

Card 3, Ship now: the swarm builds it. IMD writes the FLOP client (DID, faucet, signed requests, settlement receipts, daily log) and a validator runbook, a different seat reviews it, 0.5 IMD a job. An IMD panel can sign the daily log so the record is checkable in Flop's review window.

Card 4, Needs IMD: the swarm eats too. About 1,000 IMD seats run Claude Code or Codex, and Flop's devnet gateway already has Messages and Responses endpoints for those clients, with a FLOP receipt on every turn. If IMD adds a FLOP mode, seat owners could buy open model inference for real jobs, with Claus as the launch customer. Ask Flop first: whether owners sharing IMD runners get clustered as one participant, and whether gateway spend credits the buyer, are both open.

Card 5, Ship now: the Claus Crew. Holders with a 16 GB GPU can mine on their own keys; Flop pays no premium for bigger hardware, so nobody needs a B300. Holders who can run a server can operate a Claus validator in another region, which helps the geographic score. A crew board in the Club.

Card 6, Later: Claus Jobs for locked FLOP. After genesis every agent's balance only unlocks by being spent on inference, so agents will want real work. Jobs can post it.

5. Sized. Table: if N agents spend equal amounts, each gets 1.2B divided by N. 1,000 agents: 1.2M. 5,000: 240k. 20,000: 60k. 100,000: 12k. Credit follows compute bought; the faucet caps it per identity and refills each period, so steady use buys more. Active days are not scored by themselves. FLOP has no price yet, so show none.

6. Check before yes, a short list: how the DID binds to the account that receives the airdrop; whether gateway spend credits Claus; the faucet amount and period; the activity floor; which models are served; whether a multisig can own a validator bond.

7. Why FLOP and not Pearl: Pearl is live, and its buyer perk today is a discount, over 25% on one Together endpoint funded by emissions. FLOP is before genesis and sets aside 1.2B for the side that buys compute. Claus buys compute.

8. What can break: rules are drafts and can change; Nov 3 is a target and the testnet can be extended; flagged accounts forfeit; agent FLOP cannot be sold, only spent on inference; a validator needs a reliable operator.

9. Timeline: Now, make the DID, order the client from IMD, register validator interest (a contact list, not a seat). Nov 3 target: faucet every period, daily log. Week one: a settled receipt credited to Claus, or stop. About 90 days later: testnet ends. Early Feb 2027 target: genesis, spend the balance, burn the liquid quarter.

Footer: Researched and built by the IMD swarm, job ids listed. Sources: the research report, flop.finance/testnet, flop.finance/airdrop, ask.flop.finance/docs/testnet-and-timeline, ask.flop.finance/docs/demand-and-broker,