Borrowing Rate

The borrowing fee compensates vault depositors for the liquidity that open positions reserve. Unlike funding, which is a peer-to-peer transfer between longs and shorts, borrowing revenue flows to the protocol: it is split between the vault (liquidity providers) and the treasury. Zenex prices it with a kink model, a piecewise-linear rate on each side's vault utilization that stays cheap while the vault has headroom and climbs steeply as it fills.

Utilization

Utilization is computed per side. Each side's reserve is the value the vault must stand behind: longs mark their base tokens at the ask, rounded up, and shorts count their entry notional. The capacity each side is measured against is half the vault scaled by max_util_open:

$$\text{capacity} = \left\lfloor \frac{\text{vault\_balance}}{2} \times \frac{\text{max\_util\_open}}{\text{SCALAR\_18}} \right\rfloor$$ $$u_{\text{side}} = \min\left(\left\lceil \frac{\text{side\_reserved} \times \text{SCALAR\_18}}{\text{capacity}} \right\rceil,\ \text{SCALAR\_18}\right)$$

A side with no reserve has zero utilization. A non-empty reserve against zero capacity is full utilization (SCALAR_18). The halving floors, the capacity floors, and the quotient ceils, all pushing toward a higher rate. max_util_open is the same SCALAR_18 cap that gates new opens, where each side's reserve is checked against the same half-vault capacity.

The Kink Rate

The per-second rate for a side is piecewise-linear in its utilization u, hinged at the kink utilization target_util. Both u and target_util are raw SCALAR_18 values, so the fixed-point formulas divide by SCALAR_18. The linear leg is

$$\text{base} = \left\lceil \frac{\text{borrow\_rate} \times u}{\text{SCALAR\_18}} \right\rceil$$

Below the kink (u <= target_util) the rate is base, a gentle linear slope while the vault has room. Above the kink a steeper term is added:

$$\text{rate} = \text{base} + \left\lceil \frac{(\text{increased\_borrow\_rate} - \text{borrow\_rate}) \times (u - \text{target\_util})}{\text{SCALAR\_18} - \text{target\_util}} \right\rceil$$

Both ceils round toward a higher rate, and the rate reaches exactly increased_borrow_rate at full utilization.

The two anchor rates are SCALAR_18 per-second config values with borrow_rate <= increased_borrow_rate, both capped at MAX_BORROW_RATE (10 * SCALAR_18 / SECONDS_PER_YEAR, roughly 1000% APR). target_util is a SCALAR_18 fraction below 1.

Which Side Pays

Borrowing is charged to the dominant side. On each accrual, a side whose base tokens are strictly less than the other side's accrues nothing. The side holding strictly more base tokens advances its borrowing_idx by rate * elapsed, at the kink rate computed from its own per-side utilization. A token tie charges both sides, each at the rate from its own utilization (an empty book has zero utilization on both sides, so neither index moves). borrowing_update advances on every accrual even when no index moves.

Index-Based Accrual

Rather than touching every position on each accrual, the market keeps a cumulative borrowing index per side. On each price-bearing accrual the paying side's index advances by rate * seconds_elapsed at the rate from its own utilization (see Which Side Pays). A position snapshots its side's index at each change (borrowing_idx), and at settlement its borrowing fee is:

$$\text{borrowing\_fee} = \left\lceil \text{notional} \times \frac{\text{side\_borrowing\_idx} - \text{position.borrowing\_idx}}{\text{SCALAR\_18}} \right\rceil$$

This is O(1) per position regardless of how many intervals elapsed. Fixed-point rounding favors the protocol so it never under-collects over many intervals.

Borrowing accrual is price-bearing: it advances through accrue (which also advances funding) and at the start of every price-bearing entry point (order fills, vault-order fills, liquidations, ADL). Because of this, changing a borrowing parameter through set_config requires a same-ledger accrue, else the call reverts with BorrowingNotAccrued (703). This guarantees a rate change never applies retroactively across an un-accrued interval.