ProcessAug 21–22, 2026

Audit, price, mock, build — then write down why.

Four concepts in two days, each taken from an open‑ended idea to a usable prototype. Every decision is recorded in the design notes on each screen — press N.

  1. 01

    Audit the product, not the brand book

    OpenSea's public brand page stops at a logo and a legacy palette the live product no longer uses. So I measured opensea.io itself — CSS variables from the stylesheet, component anatomy from the DOM, layout at desktop and phone widths — and wrote it up as a 15‑section spec another team could build from.

    15‑section OS2 brand & UI specTokens as RGB triplets + a Tailwind v4 themeDo / Don't list derived from observed behaviour
    See the living system
  2. 02

    Price the idea before drawing it

    Each concept has a partner and a take. The business model shapes the screens: when the revenue is a fee, the fee box is the hero; when the risk is being classified as gambling, the credit‑default buyback toggle ships on day one.

    Partnership economics: routes, states, integration surfaceBase case ≈ $32M across four lines, with sensitivitiesCompliance treated as product requirement, not legal note
    The economics table
  3. 03

    Mock the surfaces on the real shell

    Sixteen artboards: a marketing page, the in‑app screens and the compliance‑critical states for each concept, plus the cross‑vertical home. Drawn on the real OS2 chrome so every decision met the constraints that would actually apply.

    Packs canvas · 5 artboards + business one‑pagerPartnerships canvas · 10 artboardsPack artwork system (crimped booster, tier foil)
    Browse the artboards
  4. 04

    Build the prototype so the logic is real

    One state model: a single USDC balance that margin, packs and positions all draw from, a simulated tape, and fee math shared between marketing pages and tickets so no number is typed twice. Forty‑plus routes, every edge state, mobile‑first.

    Reducer state model persisted per browserShared fee math across all three partnersDeterministic seeds so server and client agree
    Open the prototype

The economics

Take per dollar routed and reachable volume move in opposite directions across the partners. Perps are a volume business at a near‑zero rate, prediction markets are the best blend, and gacha is a high‑rate business with a small pool that matters as the on‑ramp to graded cards. Illustrative, not a forecast.

Base case ≈ $32M / yr
ConceptMechanismOpenSea takeReachable volumeRead
Packs · PSA VaultFirst‑party · PSA Vault custody1% EV margin + 5% sell‑back spread≈ $3.9B / yr categoryAn acquisition surface: 0% marketplace fee turns every pull into a listing.
Perps · HyperliquidPermissionless builder code5 bps on routed notional≈ $2.07T / yr reachableZero marginal cost, no custody. Phantom's 5 bps beat MetaMask's 10 bps on 5× the volume — price for volume.
Cards · Collector CryptBilateral share of the house edge≈ 185–200 bps of routed GMV≈ $3.9B / yr, growingThe take lands on the buyback haircut, not the pack sale — so ask for the fee pool, not GMV.
Predict · Kalshi via ApexIntroducing broker, FCM‑cleared≈ 37.5 bps (30% of a ~1.25% fee)≈ $263B / yrBest blend of rate and volume; the heaviest onboarding and most legal risk — and 62% of the base case.
Users are the shared multiplierEvery line is users × volume per user × rate. The cross‑vertical layer exists to raise the first term.
Disclose the takeEvery fee itemised before confirmation, in one component — required for supervision, and the most reused element in the study.
Eligibility before renderRestricted users never see a ticket they cannot use; the engine returns a reason so the UI can explain.
AI‑native workflow

One designer, one agent, two days.

The agent measured the site, drafted the spec, generated artboards and wrote code. I set the brief, made the product calls — the 99% EV, beating the category's 90% floor, the tier ladder, real slab photography — reviewed every screen against the system, and sent back what was wrong. Cheap first drafts move the work to deciding what the draft should be.

MeasureLive CSS variables and DOM anatomy, captured not eyeballed.
SpecifyA spec written to be executable — the Tailwind theme is part of it.
DraftArtboards on the OS2 chrome, corrected in review.
ShipPrototype walked end to end: connect, trade, pull, verify, buy.

What I'd do next

A concept is a hypothesis with a prototype attached.

Research firstFive sessions per concept on the fee box, the buyback countdown and the Kalshi ticket — where disclosure has to be understood, not just shown.
Live read‑onlyReal Hyperliquid and Collector Crypt data behind the same screens. No orders, no accounts.
Testnet tradingHyperliquid testnet with a real builder code; devnet pack purchases.
The cheapest real testHyperliquid mainnet at 3–5 bps, geo‑gated: genuine revenue data from one integration, before signing anything.
Independent concept study by Dan Weingart · Not an OpenSea product
Design systemAll concepts