The stablecoin trilemma,
solved.

01What USC Protocol Is

USC is a $1 stablecoin backed by a dual-collateral basket — 80% mETH (synthetic ETH, 1:1 backed by real ETH locked on Ethereum) and 20% WCAPE (the wrapped native token of the protocol's own chain). Every part of the system feeds one loop: real fee revenue is harvested, converted to backing, and paid out as yield to USC holders who lock their coins.

02Architecture

Two chains, one relayer, ten contracts. Ethereum holds the real ETH; the Cosmos EVM chain runs everything else.

Infrastructure

03The Five Tokens

04Contract Reference

Nine contracts on the Cosmos chain, one on Sepolia. All verified line-by-line against the deployed source.

Treasury — the mint

Issues and redeems USC against the 80/20 basket, accumulates protocol revenue as mETH, and harvests it into TCM rewards. Roles: DEFAULT_ADMIN_ROLE (deployer), OPERATOR_ROLE (deployer + MerchantBridge).

TreasuryCreditModule (TCM) v2 — the savings bank

Lock USC for 90 days at a grandfathered APR: the rate in force the day you deposit is written into your deposit and never changes for its duration. The rate offered to new deposits is a smooth linear function of the support ratio — 7% at support ≤ 1.00, 4% at ≥ 1.50 (rate = 7% − 3%·(s−1.0)/0.5). Weak peg → higher rate → more locking → active supply shrinks → peg recovers: the deposit rate is itself a peg-defense instrument. Interest remains capped by the rewards pool (solvency), and locked USC is excluded from the support denominator.

LPBonder v2 — the CAPE distributor (replaced CapeBonder)

Bond CLP tokens (not the raw basket) at a 5% discount, fixed on the deposit day — CAPE price moves during the 5-day lock change nothing. The CLP flows to the Treasury as permanent backing; equal-value USC is minted into TCM rewards. Because the deposited liquidity stays inside the pool, every bond deepens the market that must later absorb its own payout — and keeps earning the 0.3% swap fee for the protocol forever.

Payouts come from a 1,000,000 WCAPE emissions budget endowed at deploy (contracts cannot mint native CAPE — it's the gas token; the hard-capped genesis endowment is the on-chain equivalent of thin-air emissions, with the cap visible as capeReserve). Owed CAPE is reserved at purchase time, so the bonder can never over-promise.

CapePool v2 — the AMM, now with LP tokens

Constant-product (x·y=k) pool for mETH/WCAPE with a 0.30% fee that stays in reserves (k grows → CLP appreciates). Now issues ERC-20 LP shares ("CLP") with Uniswap-V2 share math, and liquidity is permissionless. Triple duty: price discovery, harvest venue, open market — plus it now mints the asset the LPBonder runs on.

MerchantBridge (Cosmos) & Merchant (Sepolia) — the bridge pair

Merchant (Sepolia) locks real ETH: depositETH() takes 0.1% in-fee, splits net 30% reserve / 70% staked (simulated 3.5% APR), and emits ETHDeposited. releaseETH() (relayer-only) pays out redemptions minus 0.2% out-fee. harvestRevenue() (permissionless) bundles accrued fees + simulated yield and emits ProtocolRevenueReady.

MerchantBridge v2 (Cosmos) is the mirror: mintBridge() (relayer-only) mints mETH 1:1 on deposits; redeemBridge() burns user mETH and signals the relayer; mintProtocolRevenue() (relayer-only) mints revenue mETH into the Treasury. The liquidity loop is now a fully autonomous, permissionless controller: executeLiquidityLoop() fires only when support < 1.02, sized smoothly to lift it to exactly 1.05 (over-collateralised model: debt ≤ 70% of pledged staked ETH — a fixed constant, like real lending terms; loan proceeds re-join staking so staged capacity converges to ~2.33× real staked); repayLiquidityLoop() fires when support > 1.30 and retires debt down to a 1.20 floor, hard-guarded by Treasury.releaseBacking(). Both have 1-hour cooldowns; the relayer checks every 5 minutes. Loop mETH is never added to totalLocked — synthetic supply can never enlarge the bridge-out allowance. The pool-seed mETH is also minted directly (not via the bridge), so bridge counters reflect only real Sepolia deposits.

Supporting contracts

05System Flows

① Bridge in: ETH → mETH

② Mint USC

③ The revenue loop (automatic, ~24 h)

④ Bond LP tokens

⑤ Autonomous liquidity loop (v2)

⑥ Bridge out: mETH → ETH

06Economic Design

The support ratio

USC locked in the TCM (deposits + rewards pool) is off the market — it cannot create redemption pressure — so it is excluded from the denominator. This has an elegant consequence: USC minted by harvestToTCM() goes straight into the TCM, so harvests raise backing while leaving active supply flat. Every harvest strictly improves the ratio.

Revenue sources

Cost of capital

The protocol "borrows" from its own TCM depositors at ≤6% APR, funded entirely by internal fee revenue — no dependency on external capital markets. This is the foundation of the Phase 2 credit facility: lend TCM capacity at 9–12% against a 6% internal cost, keeping a 3–6% spread.

CAPE flywheel — design intent and current gap

Intent: the LPBonder distributes CAPE into the market; the 20% backing rule makes every USC mint buy it back. More USC demand → more CAPE locked in Treasury → scarcer float. The v2 LP design also helps structurally: bonded liquidity deepens the very pool that must absorb bond-exit sells.

Locked design decisions

07Parameter Sheet

08P2P Fiat Ramp v2 — deployed, closed

A peer-to-peer route into USC for people who cannot use a centralised exchange. Seven contracts, live on chain 9000 since 1 August 2026, and deliberately switched off until a juror pool exists.

The rule everything derives from

Whoever can press "decline" must have something at stake. When a maker sells USC he holds that button, so his collateral headroom is the stake and the fiat buyer posts nothing at all — no deposit, no gas token, no verification, no reputation minimum. When a maker buys USC the button moves to the seller, because only the recipient of a bank transfer can know whether it arrived; so that seller escrows 120%.

The asymmetry that forces this is simple: a claim that money moved can be machine-checked against amount, timestamp, currency and rail format. A claim that money did not arrive cannot — no oracle sees an absence. Where the software gate is unavailable, capital is the only gate left.

How a trade resolves

Why griefing does not scale

Three properties do the work. Audits examine makers, not cases — a million fabricated complaints against one maker produce exactly one investigation of his whole history. The daily audit budget is an absolute number derived from bonded-maker count, never a percentage of volume, so attack volume cannot inflate the workload. And contested declines rank a maker in a priority queue, they never trigger an audit — a hard trigger would let the attacker decide how much work the system does.

Worked example: an attacker spins up 10,000 wallets against a single maker. Every trade times out or is declined, reverting instantly with no score recorded. The complaints collapse into one audit, cleared in a single pass. Platform-wide latency stays at a few days regardless of attack size — attack volume does not appear in the formula at all. To lengthen the queue the attacker must become a bonded maker himself, which costs real capital locked for a month.

The juror system

Disputes that survive the gates go to a randomly drawn, stake-weighted panel voting under commit–reveal. Panels escalate 3 → 7 → 15 → 31 → plenary, with the trade's value setting a floor the panel can never go below and randomisation extending only upward — an attacker can know the minimum but never where a case will stop.

Every level settles against the final verdict, not its own panel's majority. A juror outvoted locally but vindicated at the end is paid, not slashed. This moves the Schelling point from "what will these two say" to "what will the largest panel conclude", which is much closer to the truth.

The bound that sizes everything

Bribing a panel becomes profitable above a hard limit, so the maximum single trade is derived from the live juror pool:

max_trade  ≈  slash_fraction × (pool − top_holder_stake) ÷ 2

Sized off pool minus the largest holder, never the headline total — plenary is stake-weighted, so a single holder above half the pool would decide every case alone with no bribery required. A consequence worth stating plainly: a whale cannot buy a higher trade ceiling. Adding a 90,000 USC juror alongside ten 1,000 USC jurors moves the cap from 1,800 to 2,000, not to 20,000.

09Current Deployment (Testnet)

Cosmos EVM — chain 9000

P2P ramp v2 — chain 9000 (deployed 1 Aug 2026, closed)

Compiled solc 0.8.24, evm target paris. The v1 ramp pair — RampEscrow 0x419a…5636 and JurorArbitration 0x6816…93b2 — remain deployed but are superseded and unreferenced. See §08 for why the ramp is closed.

Ethereum Sepolia — chain 11155111

10Roadmap — The DeFi Central Bank

The stablecoin is the foundation. The endgame is a full-stack monetary system: currency, savings rate, credit facility, deposit insurance, and settlement.

Production hardening checklist