CuspCusp

Tranched Vaults

How senior and junior capital share income and absorb losses.

Vault capital is structured into two tranches with explicit seniority. The structure decides who earns what and who loses first.

The two tranches

Senior trancheJunior tranche
ForProtected, lower-volatility incomeFirst-loss capital seeking yield
Loss orderTouched only after junior is exhaustedAbsorbs losses first, in full
ReturnSteadier base incomeThe levered residual

The senior tranche is protected by structure. The junior tranche is staked first-loss capital, and its yield is compensation for taking first loss, not a bonus.

The loss waterfall

Losses flow through the structure in a fixed order. Junior absorbs first, in full, and senior is reduced only after junior is exhausted:

EJ=max(EJL, 0)ES=ESmax(LEJ, 0)\begin{aligned} E_J' &= \max(E_J - L,\ 0) \\[2pt] E_S' &= E_S - \max(L - E_J,\ 0) \end{aligned}

In one line

Losses land on junior capital first and in full; senior deposits are touched only after junior capital is completely gone.

This is standard tranche arithmetic. Income flows the other way: junior earns the levered residual once senior's claim is met.

Idle capital

Balances that are not deployed into credit do not sit dormant. Idle capital earns a base money-market rate from the first block, so depositors are paid while the protocol waits for eligible demand.

The launch bound

At launch, total credit outstanding is capped by junior capital:

iBi    EJsenior capital cannot be impaired\sum_i B_i \;\le\; E_J \quad\Longrightarrow\quad \text{senior capital cannot be impaired}

In one line

At launch, total lending is capped at junior capital, so even if every loan lost everything at the same moment, the loss could not reach senior depositors. No model or estimate is involved; it is arithmetic.

Because all borrowing is bounded by first-loss capital, even simultaneous defaults cannot reach senior depositors. Senior protection at launch rests on arithmetic, not on estimation.

The protocol scales beyond this bound only as fast as its published calibration record supports. The formal treatment is in the Cusp whitepaper.

On this page