The product
Why fully hedged
TORQUE makes one design choice on purpose: the vault is never short. Every position is bought in full as real NVDA in the on-chain pool and held by the contract until it closes. The vault lends against that NVDA; it never takes the other side of a trade. This page sets out what the other designs do, what they risk, and what we gave up for this one.
The other ways to offer leverage
| Design | Who takes the other side of the trade | What happens in a sharp gap |
|---|---|---|
| A pool as counterparty (GMX-style LP pools) | The LP pool. When traders win, LPs pay. | LPs carry the traders' net profit and loss every day, and a gap moves all of it at once. |
| Lending with liquidations (a Morpho or Aave loop) | Nobody. The lender is secured by the collateral. | The position is liquidated at the oracle price, and the liquidator's bonus comes out of the borrower's collateral. A gap past the liquidation price leaves bad debt with the lenders. |
| Order-book perps | Other traders. | Liquidations, then an insurance fund, then auto-deleveraging of winning traders. |
| TORQUE | Nobody. The position is real NVDA, held by the contract. | The position is knocked out at the first fresh print at or below its level. The vault is repaid first, and LPs lose only if a single gap goes past the worth-zero level. |
What full hedging buys
- LPs are secured lenders, not counterparties. The vault earns financing and the open fee. It neither gains nor loses when NVDA moves, as long as each loan is repaid.
- The trader's loss stops at the margin, by construction. A knock-out pays
max(sale − debt, 0): the contract never asks the trader for more. There is no liquidation bonus either; whatever is above the debt goes back to the trader. - Solvency can be checked at any block. NVDA held by the market must cover the NVDA owed to positions, 1:1. The dashboard's solvency board shows it live, and Verify every claim checks it with one command.
- There is a buffer for the weekend gap. The NVDA feed freezes for about 52 hours every weekend. TORQUE stops opening positions, and the 5% between the knock-out level and the worth-zero level absorbs a Monday gap: at 5×, the LP only loses past a single gap of about −20% (LP backtest).
What we gave up
- Capacity. Every dollar of exposure is a real purchase in the NVDA/USDG pool, which held 2.28M USDG and 3,690 NVDA at block 79,369,021. TORQUE can never offer more exposure than that pool can fill, and every open and close pays the pool's fee and price impact. A counterparty pool can offer far more exposure than any spot pool holds.
- Shorts. Hedging a short would mean holding NVDA inventory to lend, so v1 is long only.
- LP upside. The vault earns interest and fees, never traders' losses. In a market where most traders lose, a counterparty pool earns more.
- Uptime. Opens need a fresh, agreeing price, so they pause whenever the feed is more than 12 hours old: 26.6% of the last 89 days, mostly weekends.
- A paid caller. Knock-outs are permissionless but unpaid in v1 (Limits).
How big it would need to be
The demand we can measure is for credit against stocks. In Morpho's 167 USDG markets that take Robinhood stock tokens as collateral, borrowers had taken $1,456,597, 96.6% of what lenders supplied (block 78,677,903). Serving that much borrowing at TORQUE's 80% utilisation limit needs a vault of about $1.82M, or about $767k for the NVDA market alone. The $20 cap is the buildathon's limit, not the design's.
That this borrowing would move to knock-outs is our inference, not a measurement. Recompute every number here with python3 research/capacity.py.