MNX Sentinel Testnet
Litepaper, Built on MegaETH, Settled by MNX

Sentinel Predict.

A parimutuel prediction market on MNX markets, call where the AI IPO race lands, Yes or No, and get settled automatically by live MNX prices. This document explains how the protocol works, end to end.

Overview

Sentinel Predict turns MNX's AI IPO futures into simple yes/no questions anyone can take a position on. You stake a test token on an outcome; when the market closes, it settles against the real MNX mark price, and winners split the pool.

It is built by the team behind MNX Sentinel, the whale-tracking bot and dashboard for MNX, and runs fully on-chain on the MegaETH testnet. No real money is involved: it is a points-style demo using faucet tokens.

Three primitives: markets (yes/no questions derived from MNX data), parimutuel pools (stakes split by side), and an MNX price oracle (settlement by live mark price).

Why it exists

MNX lists perpetual futures on the valuations of pre-IPO AI companies. Traders already have opinions, will Anthropic cross OpenAI? will OpenAI clear $1.8T? Sentinel Predict gives those opinions a venue that is:

  • Native to MNX data. Every question is generated from MNX markets and resolved by MNX prices, no external, disputable feed.
  • Simple. Pick a side, choose an amount, done. No order book, no leverage, no liquidation.
  • On-chain & verifiable. Pools, outcomes and payouts are all contract state on MegaETH.

How it works, parimutuel pools

Each market has two pools: YES and NO. When you stake, your tokens join the pool for the side you chose. There is no fixed price, your effective odds are simply your share of the winning pool.

At settlement, the losing pool (minus a small protocol fee) is distributed to the winning side pro-rata to each winner's stake, and winners also reclaim their own stake:

payout = yourStake + yourStake × (losingPool × (1 − fee)) ÷ winningPool

If a market resolves with everyone on one side (no opposing pool), it is treated as a wash and every stake is refunded. The same happens if a market is cancelled.

Market lifecycle

  1. Create. The operator opens a market from MNX data: a question, a close time, a resolve time, and an optional cap.
  2. Stake. While open, anyone can stake tUSD on YES or NO. Live odds and the implied probability update as pools move.
  3. Close. At the close time, staking stops.
  4. Report. After the resolve time, the operator reports the outcome computed from live MNX prices.
  5. Dispute window. A short delay before funds unlock, so a bad report can be caught.
  6. Finalize. Once the window passes, the market is finalized and claims open.
  7. Claim. Winners claim their payout once; refunds work the same way for washed or cancelled markets.

Settlement & oracle

Questions are generated and settled from the MNX public market feed. Two question types ship today:

  • Head-to-head, is market A valued above market B at the close? (e.g. Anthropic vs OpenAI).
  • Threshold, is market A above a target value at the close? (e.g. OpenAI above $1.8T).

A resolver service reads the MNX mark price at resolve time, decides the outcome, and submits the on-chain report. Because the same source generates and settles the question, there is no ambiguity about what "correct" means.

House liquidity, the 1.5× match

Brand-new markets can start thin, which makes early odds jumpy and uninformative. To keep tracking effective, the protocol operator (the house) seeds depth into each market:

The house tops up its own position so that house liquidity ≈ 1.5× of the total staked by users, split across YES/NO in proportion to the current user odds, so it deepens the pool without distorting the signal.

  • Deeper pools, cleaner odds. Matching proportionally keeps the implied probability driven by real user sentiment while smoothing noise.
  • Bigger payouts. The house does not claim its winnings; its contribution flows through to the winning users, inflating payouts above a cold-start pool.
  • Testnet only. House liquidity is faucet tUSD. It carries no real value and exists purely to make the demo behave like a live book.

Fees

Protocol fee2% of the losing pool at settlement (configurable, hard-capped at 10%).
Who paysTaken from the distributed losing pool only, winners' principal is never touched.
GasPaid in MegaETH test ETH; claim from the MegaETH faucet ↗.

Test token & faucet

Stakes use tUSD, an ERC-20 test token with a built-in faucet (1,000 tUSD per claim, once per day). It is not redeemable and has no value, it exists so the market mechanics can be exercised safely.

To play: claim MegaETH test ETH for gas, claim tUSD in the app, then place a call.

Contracts, MegaETH testnet

NetworkMegaETH Testnet, chain id 6343 (0x18c7)
RPChttps://carrot.megaeth.com/rpc
SentinelPredict0x9e96C5a90b8cebcdcC4eBaa6102E67257D170149
TestUSD (tUSD)0xF807311774810d1E641e9673878761931545814A
Explorermegaeth-testnet-v2.blockscout.com ↗

Security

  • Battle-tested libraries. OpenZeppelin Ownable2Step, ReentrancyGuard and Pausable guard ownership transfer, re-entrancy and emergency stop.
  • Operator-gated. Only the operator can create markets and report outcomes; only the owner can cancel. Reports are time-locked behind the resolve time and a dispute window.
  • Solvency. Payout accounting is proven to never distribute more than a market holds; claims are once-only.
  • Tested. The core is covered by an automated test suite (payout solvency, refunds, caps, timing, fees, pause).

This is an unaudited testnet demo. Do not reuse the contracts with real value without an independent audit.

Roadmap

  • More question types (ranges, multi-outcome, time-weighted).
  • A decentralised report/dispute path beyond the single operator.
  • Shareable positions and a public leaderboard.
  • Mainnet path, pending grant support and an audit.

Disclaimer

Sentinel Predict is a community-built testnet demonstration. It uses valueless test tokens, involves no real money, and is not financial advice. Questions reference publicly listed MNX markets for settlement only.

Open Sentinel Predict →