The required DEMO-SCRIPT.md walkthrough hardcoded WQIE/TUSD addresses from the very first demo deployment; after the many redeploys today those addresses no longer exist, so the documented curl command a judge is told to "re-run before recording" actually reverted with PairNotFound. Switched it to derive the current addresses live from the /swap page (no owner-only step needed) and re-verified the exact replacement command returns the real on-chain quote. Also fixed stale/self-contradicting Foundry test counts (43/43, 41 tests) in TEAM-PRESENTATION.md and HACKATHON.md that disagreed with the real current 65/65, and the TS test count (33/33 -> 53/53, now core 23 + cli 28 + web 2).
8.1 KiB
QIE Forge — team presentation
QIE Mainnet Hackathon 3.0 · Track: DeFi & Financial Infrastructure
What we built
QIE Forge is a clean-room AMM/DEX plus an over-collateralized lending vault for the QIE blockchain, with three independent SDKs (TypeScript, Python, Rust), a CLI, and a live public frontend. It is a new financial protocol built from first principles for QIE — not a fork, not a port, not a thin wrapper around someone else's contracts.
QieForgeFactory/QieForgePair/QieForgeRouter— constant-product (x·y=k) AMM: 30bps fee compounding into reserves, minimum-liquidity burn to block the classic empty-pool attack, price-cumulative TWAP accumulators, multi-hop routing with slippage/deadline protection.QieForgeVault— over-collateralized lending: deposit, borrow up to an LTV cap, real-time health-factor computation blending QIE's real oracle with the pool's own TWAP, permissionless liquidation with a bonus, ERC4626-style lender share accounting.QieOracleAdapter— the first (and only, as far as we found) correct implementation of QIE's real oracle interface — see "Technical highlight 1" below.- Three SDKs + CLI — the same chain/oracle/AMM/vault logic implemented independently in viem, web3.py, and ethers-rs, each with its own real test suite including anvil-deployment integration tests against the actual compiled contract bytecode.
- Live frontend — https://qie-forge.vercel.app, public, no login, real on-chain data on every request.
Technical highlight 1 — we found and fixed a bug in QIE's own documentation
QIE's public Oracle docs (docs.qie.digital/qie-oracle) specify
function getPrice(string) view returns (uint256). We called that exact
signature against the real deployed BTC price feed on QIE Mainnet. It
reverts — the function doesn't exist on the deployed contract. By decoding
the actual bytecode's function selectors, we found the real deployed
contracts are Chainlink AggregatorV3Interface clones (decimals(),
description(), latestRoundData()), confirmed with live eth_calls, not
guesses. QieOracleAdapter and all three SDKs implement the corrected
interface and fail closed rather than trusting the docs. Most hackathon
entrants integrating QIE's oracle for the first time would copy the
documented (wrong) signature and either silently fail or fabricate a result —
we found the real contract behavior and built against it.
We also found and fixed: the documented testnet RPC endpoint
(rpc1testnet.qie.digital) is dead (ECONNREFUSED) — still listed on the
hackathon site's own "Builder Quickstart" today — and the real working
endpoint is QIE's Blockscout explorer's own JSON-RPC proxy; and that the
oracle contracts only have bytecode deployed on Mainnet, not Testnet, at the
same addresses.
Technical highlight 2 — a real security bug found and fixed pre-deployment
An external security reviewer flagged a real issue in QieForgeVault. We
independently re-verified it by reading the contract source directly, not
taking the report on faith:
- No lender deposit/withdraw function existed — liquidity could only enter via a direct token transfer, with no path for a lender to ever reclaim principal.
liquidate()could be gamed to leave permanent bad debt: a liquidator could choose the smallest repay amount that still exhausted the available collateral, leaving debt principal stranded against zero collateral, forever, with nobody incentivized to ever callliquidate()again on that position.
Both were fixed (commit f55e3c6) before any on-chain deployment — no funds
were ever at risk. We added ERC4626-style lender share accounting priced
against totalAssets(), and capped the liquidator's effective repay amount
to what the seized collateral actually covers, writing off the shortfall via
a BadDebtWrittenOff event instead of leaving a dangling balance. A
dedicated regression test,
test_liquidation_minimalRepayStillZeroesDebt_whenCollateralExhausted,
reproduces the exact exploit path and proves debt and collateral now always
reach zero together. We then ran the same adversarial scrutiny over the
AMM/Router (reentrancy, TWAP manipulation, LP rounding, access control,
slippage/deadline enforcement) and found no further exploitable issue —
documented honestly either way, not just when we found something.
Technical highlight 3 — verified, not just tested
- 65/65 Foundry tests across the full contract suite.
- 53/53 TypeScript (core 23 + CLI 28 + frontend 2), 23/23 Python, 15/15 Rust tests, each SDK built independently (not a thin wrapper around the others).
- Real anvil-deployment integration tests in both the Python and Rust SDKs: they deploy the actual compiled contract bytecode (not a mock) to a local devnet and prove the SDK's computed health factor and swap quote match the deployed contract's own values exactly, including a built unsigned transaction that actually executes on-chain.
- A real, persistent, publicly reachable demo deployment — a self-hosted anvil devnet with the real compiled contracts deployed, seeded with real liquidity and an open vault position — so the live frontend shows real on-chain reads end-to-end today, while QIE Testnet/Mainnet deployment waits on a funded deployer key.
Judging criteria mapping (QIE's published weights)
| Criterion | Weight | QIE Forge |
|---|---|---|
| Innovation | 30% | Clean-room AMM + lending vault, built around an independently-discovered oracle-docs bug most entrants would copy blindly. |
| Technical Quality | 25% | 65/65 contract tests, 3 independent SDKs with real anvil-integration tests, six real security bugs (liquidation bad-debt, lender-share donation/inflation attack, lender-share pricing staleness, a bonus-rate inconsistency in the liquidation-curve work caught in the same pass, a bad-debt-write-off-vs-realized-protocol-reserves insolvency/DoS gap that could permanently freeze every lender's funds, and a deposit() division-by-zero panic that permanently blocked new lending liquidity after a full bad-debt wipeout) found and fixed pre-deployment, each with a reproducing regression test. |
| User Experience | 15% | Live public frontend, real swap/vault flows working end-to-end, no login wall. |
| Mainnet Integration | 15% | Oracle/RPC/explorer integration is real and QIE-Mainnet-verified today; the protocol's own contract deployment to QIE Mainnet is the one open gap (see below). |
| Business Potential | 15% | A DEX + money market is the load-bearing DeFi primitive QIE doesn't have yet, and the vault now has a real on-chain revenue line (reserveFactorBps, a configurable cut of borrower interest withdrawable by treasury) plus a severity-scaled liquidation incentive curve (liquidationBonusMaxBps) that reduces expected bad-debt losses against that revenue — on top of the recurring swap/LP/borrow activity the hackathon's traction milestones reward. |
What's honestly still open
- QIE Testnet/Mainnet deployment of the protocol contracts. QIE's FAQ states plainly: "Testnet-only projects are not eligible. Judges verify contracts, transactions, deployment, and frontend connectivity on QIE Mainnet." This is blocked on a funded QIE Mainnet deployer wallet — a real owner/operator step (QIE's testnet faucet is a connect-a-wallet flow, not a plain API; mainnet needs real funded QIE). Everything else is built and ready to point at it the moment it exists.
- No real-browser wallet click-through test (needs an actual wallet extension).
- QUSDC integration stays fail-closed — its real contract address was never independently confirmed publicly.
- QIE ID is not integrated — no public API or contract was found for it.
The team
Built solo for QIE Mainnet Hackathon 3.0, with AI-assisted development (QIE's
own FAQ: "Yes, AI-assisted development is encouraged. Teams must fully
understand and maintain their code and be able to explain architecture to
judges.") — every claim in this document and in docs/STATUS.md is backed
by a command you can re-run yourself, which is the standard we held
ourselves to throughout.