Forge

Got a question? Make it a market.

Forge deploys HIP-4 outcome markets on Hyperliquid testnet. You pick a validator-approved question format, fill in the blanks, sign with your wallet, and your question goes live on a real order book where anyone can trade yes against no. When the deadline passes, you post the result and the winning side redeems at $1 per token.

What's in the app

Create. Nine question formats, read live from the on-chain template registry: price targets (mark price, candle close, or TWAP-averaged), price touch, price ranges, sports match winners, and over/unders on any verifiable number. The wizard shows human field labels, validates as you type, converts your local time to UTC and tells you so, and renders a live ticket of exactly what will go on chain. No raw {placeholder} syntax anywhere.

Live markets. The full testnet board with search, sort (newest, oldest, A to Z), and category filters parsed from the machine metadata deployers embed in descriptions. Testnet is full of throwaway deploys named template:whatever; Forge hides them behind a toggle and ranks real questions first. Click any market for a detail view: implied odds for both sides, a four-level depth ladder from the live order book, the spread, and the full resolution criteria.

Settle. Markets you deployed appear under Mine. Settlement is a two-step armed confirmation, disabled until the market's settlement time, with a price check that reads the 1-minute candle at the settlement timestamp (not the current price) once the market is due. Every check carries evidence: the source, the value, and a sha256 of the raw response.

Running it

npm install
npm run build
npx next start -p 5210

Optional .env.local:

Variable	What it does
FORGE_QN_TESTNET_ENDPOINT	Quicknode Hyperliquid testnet endpoint. Info reads route through it server-side; the token never reaches the browser.
FORGE_QN_ENDPOINT	Quicknode Hyperliquid mainnet endpoint, used by the settlement price oracle.
NEXT_PUBLIC_SITE_URL	Public URL for OpenGraph metadata.

Without the endpoints, reads fall back to the public Hyperliquid API.

What deploying actually requires

This runs against Hyperliquid testnet. To deploy markets your wallet needs:

• 100 testnet HYPE staked as an outcome deployer
• At most 10 active outcomes at a time
• At most 50 deployments a day

Settlement is the deployer's job and wrong settlements are slashable. The app repeats this in the UI because it matters.

How signing works

Hyperliquid L1 actions sign against a phantom EIP-712 domain: keccak(msgpack(action) || nonce_u64_be || 0x00) becomes the connectionId of an Agent{source, connectionId} message on chainId 1337, where source "b" means testnet. Your key never leaves the wallet; the app sends only {action, nonce, signature}.

One sharp edge worth knowing: viem 2.5x refuses to sign typed data whose domain chainId (1337) differs from the connected chain (HyperEVM testnet is 998). Forge bypasses this by calling eth_signTypedData_v4 on the connector's raw EIP-1193 provider (lib/useL1Sign.ts), which wallets accept without a network switch. This is also how Hyperliquid's own frontends do it.

The HIP-4 action schemas

These are not officially documented anywhere. They were read back from the live API and verified against community SDK source:

// deploy a standalone outcome from a template
{
  "type": "spotDeploy",
  "outcome": {
    "registerStandaloneOutcomeFromTemplate": {
      "id": "binaryPrice2",
      "keywordToValue": [["perp", "BTC"], ["threshold", "80000"], ["time", "20260901-1200"]],
      "deployerFeeScale": "0"
    }
  }
}

// settle it ("1" = first side wins, "0" = second side)
{
  "type": "spotDeploy",
  "outcome": {
    "settleOutcome": {
      "outcome": 10218,
      "settleFraction": "1",
      "details": "Settled per price read ...",
      "nameAndDescription": ["...", "..."],
      "sideNames": ["Yes", "No"]
    }
  }
}

keywordToValue must be sorted by byte order. Outcome asset ids follow #(10 * outcomeId + side); that string works in l2Book and order placement. Prices live in [0.001, 0.999], sizes are integers, collateral is USDH/USDC depending on the market.

Data plumbing

All chain reads go through two API routes so Quicknode endpoint tokens stay server-side. The routes constrain request shape, not volume: they carry no auth or rate limit of their own, so put a rate limit in front of them before hosting publicly (see Deploying).

• POST /api/info rebuilds the upstream body from a per-type whitelist (outcomeTemplates, outcomeMeta, meta, l2Book with a coin) and rejects everything else. Bodies over 512 bytes bounce.
• POST /api/oracle reads either the current mid (allMids) or, given a past atMs, the 1-minute candle at that timestamp. The response says which basis it used, and the UI treats current-price reads as indicative only.

Design

The visual system is a small set of --qn-* semantic tokens with oklch values, Geist for text and display, Geist Mono for numbers, elevated surfaces as 5% overlays, and appearance that follows the OS. Forge's own layer on top: yes is green, no is purple, everywhere and always. The wordmark's "o" is a disc split between the two. Tickets get punch-hole perforations and a stamp. The question-mark glyph (a drawn hook whose dot is the split disc) appears large in the hero and small in empty states.

Screenshots:

Create	Board	Market detail

Known gaps

• "Mine" is per-browser localStorage. Clear it and the settle UI for your markets is gone until you use "Find market id". Chain-side reconciliation by deployer address would fix this.
• There is no known info endpoint to check deployer stake before the first deploy attempt. The error message handles it after the fact.
• Question-type markets (multi-outcome, like rate decisions) can be browsed but not yet created; the wizard covers standalone outcomes.

Structure

app/            pages, API routes, metadata, icon, OG image
components/     TemplateGallery, MarketWizard, Ticket, LiveMarkets, MarketModal, QMark
lib/            hl.ts (actions, info client), sign.ts (L1 hashing), useL1Sign.ts,
                templates.ts (humanized registry), errors.ts

Testnet only. Settle honestly.

Deploying

The app is a plain Next.js project: one static page and two Node API routes. On Vercel, set these in Project Settings:

Variable	Value
FORGE_QN_TESTNET_ENDPOINT	Quicknode Hyperliquid testnet endpoint URL
FORGE_QN_ENDPOINT	Quicknode Hyperliquid mainnet endpoint URL
NEXT_PUBLIC_SITE_URL	The production URL, for OpenGraph metadata

Both routes wait up to 20 seconds on upstream, so set the function max duration to at least 25 seconds.

/api/info and /api/oracle are unauthenticated and every request spends credits on your endpoints. Add a rate limit before sharing the URL. On Vercel: Firewall, Rules, New Rule, condition Request Path starts with /api/, action Rate Limit, 60 requests per 60 seconds keyed by IP, then Deny on exceed. Adjust the numbers to your plan.

License

MIT, see LICENSE.

The Geist and Geist Mono font files in public/fonts/ are copyright Vercel and distributed under the SIL Open Font License 1.1; the full text is in public/fonts/OFL.txt.