LEVER: Perpetual Leverage on Prediction Market Probabilities

A trader who believes an event has a 70% probability must pay $0.70/share for a max return of 43% — vs. 10–100× leverage available in traditional derivatives. This structural absence of leverage suppresses sophisticated participation, weakens price discovery, and constrains volume.

LEVER's solution: a synthetic leverage protocol that references external prediction market prices, with its own margin, liquidation, and settlement infrastructure.

Five core innovations

Fee distribution (Tier 1, IFR < 20%)

The vAMM breakdown problem

Constant-product vAMMs maintain x · y = k. Near probability 0 or 1, one virtual reserve approaches zero, producing catastrophic slippage exactly where leverage is most valuable.

LEVER's linear impact is independent of PI level — a $1K trade at PI=0.95 incurs the same base impact as at PI=0.50.

16 components

Three core principles

Without smoothing, raw prediction market prices would cause erratic margin checks, false liquidations, and manipulable PnL. The Smoothing Engine produces four output properties: continuity, stability, monotonic convergence, and anti-manipulation resilience.

Four sub-systems

Combined update formula

Six required properties

PnL calculation

Entry price ≠ Mark price. Entry comes from the execution model (with impact). PI determines PnL, margin, and liquidation from that point on.

Concentration Factor (M_conc)

If a market exceeds 15% of global OI, the concentration factor activates — driving down leverage and driving up borrow fees & margin to choke off toxic concentration.

Two curves, one formula

R(τ) with τ_ref = 24h drives mechanical constraints (leverage, MM, OI caps, execution depth, oracle freq).

R_borrow(τ) with τ_ref_borrow = 168h drives borrow rate escalation — starts a full week earlier than mechanical tightening.

The three-act structure

The dual-curve design creates a natural narrative across the position lifecycle:

Growth tier summary

Why this beats vAMM

• No boundary breakdown: impact is independent of PI level. A $10K trade at PI=0.95 has the same percent impact as at PI=0.50.
• Balance incentive built in: trades that improve balance receive reduced impact (down to 0). Trades that worsen balance pay more.
• No virtual reserves, no resets, no drift management. Stateless except for current OI balances.

The pincer effect

Near resolution, two forces squeeze equity simultaneously:

• MM rises from below: as R(τ) drops, MM_Multiplier climbs from 1.0× → 3.0×.
• Borrow fees erode from above: M_ttR escalates from 1.0× → 25.0×, draining equity continuously.

The lines converge faster as resolution approaches because both forces accelerate. This is the mechanism that forces positions to close before resolution.

Grandfathering

When caps compress (R drops, market goes live), existing positions are never forcibly closed. The cap acts as a one-way gate: once OI exceeds the new cap, only reductions are permitted until OI falls back below.

Five liquidation triggers

1. Adverse PI movement — market moves against position.
2. Borrow fee erosion — continuous accrual drains equity.
3. MM rise from R(τ) compression — pincer effect (bottom jaw).
4. Funding rate drain — heavy-side traders pay funding.
5. Combined vectors — all of the above simultaneously, common during LIVE events.

Three execution paths

Bad debt waterfall

Matched vs unmatched split

Matched OI = min(longOI, shortOI). Funding on this portion is trader-to-trader, zero-sum.

Unmatched OI = |longOI − shortOI|. Funding on this portion flows directly to the LP pool as counterparty risk compensation. Funding is not protocol revenue — it bypasses the 50/30/20 split entirely.

Zero-imbalance: when longOI = shortOI, imbalance_ratio = 0 and funding_rate = 0 — no one pays, no one receives. The LP pool carries no net directional exposure, so no compensation is owed. This is the equilibrium the system pushes toward.

Funding interval: accrual is continuous but settles on an interval that tightens from 1 hour (far out) to a 5-minute floor (at resolution): max(300, min(3600, 3600 · R_adjusted)) seconds. The · 4.0 above is the funding_escalation constant (max 5× at resolution).

1× positions are exempt: at full collateralization, the trader isn't borrowing LP capital. Charging borrow on fully-collateralized positions would be economically incoherent and would discourage participation exactly when the system wants positions open for price discovery.

Imbalance surcharge: additive premium on the heavy side. + imbalance_ratio · 1.0. At 100% imbalance, the heavy side pays 2× the light side's rate.

Why two curves, not one

Four revenue sources

Revenue projections by platform stage

Assumes monthly volume ≈ 15× TVL at an all-in ~0.20% take rate.

Dual-contract architecture

The separation is deliberate: NAV fluctuates with trader PnL (principal risk), while yield accrues steadily in a separate contract (claimable without burning shares, without cooldown, without affecting utilization).

NAV & share price

Unrealized trader PnL is included deliberately: if NAV ignored it, an LP could withdraw at an inflated share price while traders hold large unrealized profits — leaving the remaining LPs to absorb the loss when those positions close. Including it makes every deposit and withdrawal NAV-neutral. lvUSDC is the ERC-4626 share token; its price floats above or below $1.00 with pool performance.

Six-layer risk mitigation

1. OI Caps — limit total directional exposure.
2. Funding — real-time compensation for imbalance risk.
3. Borrow Fees — force position closure before resolution.
4. Leverage Compression — positions reach 1× at resolution.
5. Liquidation Engine — force-close underwater positions.
6. Insurance Fund — absorbs bad debt before LP pool.

Withdrawal mechanics

• 48-hour cooldown after deposit — prevents timing attacks around known fee-generating events.
• Utilization gate: withdrawal blocked if it would push global utilization above 80%.
• Reward claims bypass both — fees already earned, no reason to gate.

The 80% gate produces natural scaling — withdrawable capacity shrinks as utilization rises:

Market state transitions

During PENDING_RESOLUTION (the oracle gap): no new positions, no voluntary closes, fee accrual frozen at external timestamp, liquidations continue with a 2× MM multiplier (MM_at_resolution × 2 = 2.5% × 3.0 × 2.0 = 15% of notional) to clean up near-zero-equity positions before settlement.

Settlement payout

Outcome PnL is positive for winners (longs on YES, shorts on NO), negative for losers. If a winner's equity is positive but below the settlement fee, the fee is reduced to the remaining equity — never a negative payout. If a loser's final_equity ≤ 0, payout is $0 and the deficit enters the bad-debt waterfall.

Four-layer bad debt waterfall

Three insurance constraints (pay the MIN)

ADL haircut (Layer 3) covers whatever insurance can't:

Applied pro-rata across that market's winners only (market-isolated). Every winner takes the same percentage haircut and still profits — just less.

Fee accrual freeze

Borrow & funding accrue up to the external resolution timestamp (when Polymarket/Kalshi declared the result) — not the on-chain recording timestamp. Charging fees during the oracle gap, when the outcome is already known and the LP pool bears zero risk, would be economically incoherent.

Void / cancellation

If the source platform cancels the market, positions refund at current equity with no settlement fee. Accrued fees are not reversed — they paid for real risk borne while the position was open — and there is no bad debt on void (the LP pool already received those fees). Zero or negative equity refunds $0.