Options on the borrow rate

The HyperCore USDC borrow rate floats, and nothing on Hyperliquid lets you fix it. The track built an index on the rate and a ladder of HIP-4 scalar bands across its kink: buy a band to fix part of a cost, sell one to lock a yield. The deciding test, listing on a perp somebody else deployed, passed.

The rate is 0.05 + 4.75 * max(0, u - 0.8) a year on utilisation u, accrued continuously and indexed hourly. Past 80 per cent, each point of utilisation adds 475 basis points. Every borrower, including portfolio-margin accounts that borrow automatically, is short a call on utilisation.

Say utilisation rises from 80 to 81 per cent:

• The rate goes from 5.00 to 9.75 per cent.
• The index, quoted as 100 - rate, falls from 95.00 to 90.25.
• Settled there, the 90 to 95 band pays Long (90.25 - 90) / (95 - 90) = 0.05 and Short 0.95 per contract.

The index is quoted as 100 minus the rate

This is the rate-futures convention: 5 to 25 per cent maps to 95 to 75. A HIP-3 rule forced it. An externalPerpPx move over 50 per cent from the start-of-day price triggers a validator slashing review, and assets expected to do that more than once a month cannot be cross margined. Through the kink, 5 to 10 per cent is +100 per cent as a rate but 5.3 per cent as 95.00 to 90.00.

• It tracks the hourly-indexed realised accrual, not an average of the instantaneous rate, so a hedge is exact.
• It recomputes the rate from utilizationBps rather than trusting borrowYearlyRateBps, so divergence is detected. Five gates fail closed on degenerate inputs.

Each band is a fully collateralised call spread

No liquidation, no auto-deleveraging. A scalarPrice Long pays (price - low) / (high - low) clamped to 0 to 1 and Short pays 1 minus that, on a TWAP of the named perp over a stated window. Adjacent bands replicate a payoff curve far better than as many single strikes.

A Long band is short the rate: borrowers hedge with Short, suppliers lock a yield with Long. The kink at 95.00 is steep: the 90 to 95 band covers rates of 5 to 10 per cent, utilisation of 80 to about 81.05 per cent. See market shapes.

The template accepts somebody else’s perp

The project hinged on one keyword. scalarPrice names its underlying through hlPerp; if it rejected a builder-deployed dex:COIN, only a rate perp would remain. The docs hinted : was allowed, but mainnet had 0 scalarPrice outcomes among 228 active.

Proven by the track’s own transaction:

Evidence	Finding
Shared type	scalarPrice, binaryPrice and priceTouch all declare perp as hlPerp on mainnet; validation is per keyword type
Already used	35 of 228 live mainnet outcomes carried a dex-prefixed hlPerp value
Own markets	Four testnet scalars on a builder-deployed US 10-year Treasury yield perp, accepted first time and read back from outcomeMeta: bands 4.5 to 4.75, 4.75 to 5, 5 to 5.25 and 5.25 to 5.5, expiring 2026-10-15 at 16:00 UTC, on the 3,600-second TWAP
Testnet template	scalarPrice3 instead of scalarPrice, with the same keywords, name and side names

An outcome deployer need not own the perp dex: all three mainnet outcome venues on the largest builder-deployed dex were third parties to it.

uDecimal values forbid trailing zeros: 95.000 signs and prints normally but is refused on submission; 95 is accepted.

Moving the index costs only opportunity

Smoothing sets how long an attacker must hold, not whether they can. Routes to the kink against the 2026-09-20 reserve:

Route to the kink	Capital	Marginal cost to hold 8 hours
Borrow until utilisation reaches 80%	55.8 million USDC borrowed, plus 87.3 million USD of HYPE collateral at 0.65 LTV	about 3,110 USD of interest
Borrow further, to a 6.00% rate	56.8 million USDC	9,300 USD of borrowing per extra basis point
Withdraw own supply until utilisation reaches 80%	69.8 million USDC	about 1,958 USD of forgone yield
Withdraw further, to a 6.00% rate	70.8 million USDC	9,788 USD of withdrawal per extra basis point

Withdrawal fits one account under the 250 million USDC per-user supply cap; borrowing needs two (50 million USDC per-user borrow cap). A holder of Q dollars of a bounded side gains at most Q, so with hurdle h on committed capital C held for t, and safety multiple k:

Q_max = C * h * t / k

At a 20 per cent hurdle and 3x safety, defending 10 million USD needs over 900 days. The real lever is reserve depth, which no deployer controls and which is thinnest near the kink, where the capital to cross 80 per cent collapses towards zero.

Owning the perp does not pay

The HIP-3 stake cannot be earned back in fees. 500k HYPE, held at least 183 days after the dex is deployed and burned if validators vote to slash, needs the month-defensible book to turn over 187 times a day; a liquid perp turns over a few times. No row of the cost schedule (HYPE 45.69 to 182.75 USD, hurdles of 5, 10 and 20 per cent) closes the gap, even with slashing unpriced.

The testnet deploy failed on collateral, not stake: with 205 HYPE staked, registerAsset2 returned Invalid perp DEX for USDC and Collateral asset not supported for five other tokens. The deploy auction stood at 500 HYPE against 0.4 HYPE on the account.

The numbers

Quantity	Value	Basis
Borrow rate curve	0.05 + 4.75 * max(0, u - 0.8) a year	docs; spec 01
Rate at 80%, 90%, 100% utilisation	5.00%, 52.5%, 100%	computed from the curve
Supply rate identity	borrow * u * 0.9; 500 x 0.6736 x 0.9 = 303 bps	spec 01; spec 02
Live reserve, 2026-09-20	5.00% borrow, 3.03% supply, 67.36% utilisation; 441.7 million USDC supplied, 297.6 million borrowed	spec 01
Index quote	100 - rate; kink at 95.00	spec 01
Mainnet templates, testnet templates	18 and 66; scalarPrice on mainnet only	spec 01, 2026-09-20
Live outcomes with a dex-prefixed hlPerp	35 of 228	spec 01
scalarPrice outcomes on mainnet	0 of 228	spec 01
Defensible open interest, 8-hour window	3,296 USD	spec 01 (20% hurdle, 3x safety)
Defensible open interest, month window	296,677 USD	spec 01
HIP-3 stake	500k HYPE, at least 183 days, burned if slashed	docs
Stake at 91.25 USD a HYPE	45.6 million USD	spec 01
Carry at a 10% hurdle	4.6 million USD a year	spec 01
Breakeven volume, 0.045% fee, 50% deployer share	55.6 million USD a day	spec 01
Turnover of the defensible book to break even	187 times a day	spec 01
HIP-4 mainnet limits	100 active outcomes, 500 new a day	digest 02

Keep the bands, drop the perp

Listing on an existing perp removes the HIP-3 stake, oracle liability and slashing, but makes you a HIP-4 deployer: 500k HYPE on mainnet unless you list from an existing venue, and the duty to compute the TWAP and submit it with exactly matching names.

A rate underlying already existed, contrary to the brief: a US 10-year Treasury yield perp on mainnet, quoted as the yield in percent, with open interest. Every rate and volatility underlying found had cross margin off.

History has to be collected. There is no historical 0x812 reserve state, so a sampler records the reserve on a fixed grid while the index is tested on synthetic and adversarial series.

Orders are not delayed either, contrary to the docs: see one HyperEVM transaction.

Six questions still open

Open question	Detail
The underlying is not the HyperCore rate	No borrow-rate perp exists, so the cheap path uses an external rate such as a Treasury yield
A halt on somebody else’s perp	An oracle the lister does not control; its effect on live bands is not written down
Validator acceptance of a rate as a HIP-3 underlying	Slashing guidance does not reach index choice, and no probe returns a stake-weighted judgement; matters only if the perp is revived
Reflexivity	Whether a borrow-rate perp changes borrowing is unmodelled, and a fair reason for validators to decline
Real history	The smoothing window cannot be validated until the sampler has run several multiples of it
What would flip the verdict	A much lower HIP-3 stake (the HIP-3 page expects a decrease; at 50k HYPE breakeven falls tenfold), or an order of magnitude more depth in the HyperCore USDC reserve

For the hlPerp bridge, see How the HIPs compose; for stake cost and slashing, Deployer economics and risk.