Skip to content
Request a review
Selected markets from the xyz and km HIP-3 dexes
Hyperliquid public API ·
Fabrknt for HIP-3

Day One: a launch pool for new HIP-3 listings

For the deployer of a new HIP-3 listing: a counterparty pool that keeps a thin book tradable from its first day, with limits aimed at the attacks thin books invite, each checked against a named replay or test. A reference implementation the deployer runs, not a product.

Testnet onlyNot audited

What it studies

01

Initial liquidity

A counterparty pool the operator seeds and traders' losses and fees grow. It takes a new position only while a set move either way (10% on testnet, never under 5%) would still leave it able to pay every winner, and is retired once the book is deep enough.

02

Price anomalies

Fills and liquidations wait whenever HL Price Guard rejects the price: a jump since the last accepted price, a gap to the book, or a gap to an outside reference.

03

Exposure caps

Opens and closes each have a rolling cap per minute, sized from the book's depth, and net open interest is capped.

04

Halted markets

When a listing is halted or delisted, the book goes empty and every price check refuses. Once the oracle has stayed still for an hour, positions can close and the pool can pay out at that price; the operator can also end the pool at the price the market settled at.

05

Who carries the losses

Losers get pool shares worth 12% of what they lost. Winners leave part of their profit in the pool: 30% on small moves, falling to 15% once the price has moved 1%, as Papertrade does. The rebate is kept below even the lowest haircut, so trading against yourself from two accounts loses money at any size. Rebates are not cut for large losses either: the pool lives only until the book is deep enough, so the dilution has an end.

What it found

2026-10-10 · Papertrade

Replaying the 228 positions: settling at the oracle instead of the book cut the take by only 15%, because the oracle moved with the push. An exposure cap of $3.3M per minute would have cut the take to an estimated $55K, below the $130,873 the push cost (assuming the incident's average edge and pushes that cost what Papertrade's did; an estimate, not a bound).

2026-07-27 · SK Hynix

After the first bad oracle update (-15.6%), the price stepped down 1% at a time. A check against the previous price lets 39 of those updates through; holding the last accepted price lets none through, and pauses for 2 min 12 s.

Safe caps · 2026-10-19

Measured from the book over a full US session (2026-10-19), the safe settlement rate was about $50K a minute for xyz:TSLA ($40K in the opening hour) and $20K for xyz:SKHX. On the day of its bad update (2026-07-27), xyz:SKHX carried $376M of open interest.

Status

  • Deployed on HyperEVM testnet against xyz:TSLA, with test USDC: pool 0x4B0d8AB73b2511EbDF9eBE474cf0C5a2a77AD8cD, seeded with 10,000 (2026-11-19). A keeper pushes Hyperliquid mainnet's oracle for the same market to it every minute as a reference price, for testnet only. Not deployed to mainnet, and not audited. It is not offered to traders.
  • On the previous pool (2026-11-14), the first position, a $200 long, waited while the testnet book had a 378.38 bid and a single 500 ask, a 28% spread against a 1% bound; it settled on 11-16 once the spread narrowed, at 380.96. But the testnet oracle had stopped at 380.96 on 11-14 while mainnet's moved on, and with nothing to compare it with, the check could not tell. The current pool adds the reference: if the oracle is more than 1.5% from it, or it goes stale, settling waits. Its first position, a $200 long requested on 11-19, is waiting, because the testnet book has bids and no ask. Watch the pool as it is now, with no wallet: the prices, the check's verdict and what the pool can pay.
  • It uses HL Price Guard pinned to a tagged version (v0.4.1). A deployer can run the whole of it locally with no keys (npm run demo) and size the seed, caps and latch for their own market with its sizing guide. The source is public under Apache-2.0: fabrknt/dayone.
  • It moves forward when an operator has a concrete need for it, for example a new or pre-IPO listing whose book will start thin; if that is you, request a readiness review.

How it was checked

1

Thirteen numbered rounds of review by a separate AI model (Codex), each with follow-ups, across the contract, the keeper and its installer; the thirteenth found nothing new. Five more rounds covered the testnet reference price and the keeper's push to it; the last of them left one finding recorded rather than fixed: if the keeper restarts while a push is still unmined, its next transaction can use the same nonce, and one of the two is rejected and retried on a later tick. This is not an audit. Each finding was first shown with a test that passed only because the flaw existed, then fixed, then the same test was run again.

2

The reviews found and closed accounting and share-pricing flaws, among them a way for one trader's open loss to offset another's open profit, ways to take value from earlier shareholders by trading against oneself, a way to collect shares just before pending orders' fees reached the pool, and a way to pair an old haircut with a newer, higher rebate across a parameter change. Shares are now priced at the most the pool can reach from the positions open or requested at the time, and redeemed at a conservative value.

Operator Readiness Review

For teams opening a HIP-3 market, or already running one. We check your setup against what has gone wrong for HIP-3 operators in the public record, and say what we could not check.