INITIALISING SIGNAL ENGINE
APEX PREDATOR · POLYMARKET · LONDON

HUNT. STRIKE. WIN.

In a jungle of predatory markets, MANTIS waits perfectly still. Pre-screen filters the noise, Oracle locks the edge. One precise strike.

0%
Win Rate
0
Markets Backtested
0
Active Strategies
GEO-UKRAINE-Q2
FED-RATE-JUNE
EU-GAS-BAN-Q2
BTC-100K-MAY
ORDER PLACED
Scroll
WIN RATE71.6%
MARKETS BACKTESTED194
PRE-SCREENnoise filtered · fast-pass gate
ORACLE CONSENSUSdual-stage pipeline · deep verify
MAX EXPOSURE$40
DAILY LOSS LIMIT−$15 / day
ORDER LATENCY< 50ms to CLOB
RUNTIMERust · Tokio · London EU
CAPITAL$100 USDC
ACTIVE STRATEGIES4 independent execution paths
TAKER FEE1% standard · 0% geopolitical
WIN RATE71.6%
MARKETS BACKTESTED194
PRE-SCREENnoise filtered · fast-pass gate
ORACLE CONSENSUSdual-stage pipeline · deep verify
MAX EXPOSURE$40
DAILY LOSS LIMIT−$15 / day
ORDER LATENCY< 50ms to CLOB
RUNTIMERust · Tokio · London EU
CAPITAL$100 USDC
ACTIVE STRATEGIES4 independent execution paths
TAKER FEE1% standard · 0% geopolitical
Performance Snapshot

Numbers that don't lie.

Win Rate
0
Across 194 backtested markets — validated against real Polymarket resolution data
Markets Backtested
0
Politics, Crypto, Sports, Economics, Geopolitical, Tech and more — all categories covered
Daily API Cost
~$0
Prompt caching cuts Oracle input costs significantly. Pre-screen stage runs at minimal cost. Pipeline optimised for capital efficiency.
Latency to CLOB
0
London execution node co-located with Polymarket's matching engine for sub-5ms connectivity
Signal Generation

The Signal Pipeline.

Two AI stages in the normal path. Each eliminates noise before the next fires. Only the highest-confidence signals reach execution — edge maximised, capital protected.

PRE
Pre-Screen Fast-Pass
Low-signal markets filtered
Sub-100ms inference. Acts as the cost and latency gate — eliminates low-signal markets before the expensive deep-verify stage runs. Scores question clarity, price edge, and time-to-resolve. Only strong candidates advance to Oracle.
Fast-pass · Stage 1
LPU Inference<100msNoise FilterRate-limited
ORC
Oracle Deep-Verify
High conviction threshold to trade
The primary decision stage. Carries dominant consensus weight. Applies category-aware reasoning rules and fee awareness. Consensus combines Oracle and Pre-Screen scores. Prompt caching keeps input costs minimal. Normal path: Pre-Screen → Oracle → Risk Gate.
Deep-verify · Stage 2 · Primary
Primary WeightPrompt CachingCategory RulesWeighted Consensus
RES
Resilience Layer
Emergency fallback — Oracle offline only
Activates automatically only when the Oracle stage is unreachable. Not part of the normal execution path. Provides category-aware context and domain base rates as a substitute. Ensures zero signal loss during outages — the pipeline never fully goes dark.
Emergency fallback · Not in normal path
Auto-ActivateOracle Offline OnlyZero Signal LossFail-Safe
EXEC
Risk Gate → CLOB Execution
8 rules · Kelly sizing
Every passing signal clears 8 hard risk rules before any order reaches the CLOB. Fail-closed — a single violation drops the trade entirely. Kelly-based sizing scales per-market risk dynamically. No exceptions, no workarounds.
Polymarket CLOB REST + WebSocket
Kelly Criterion8 Risk RulesFail-Closed<50ms Order
Validation

Backtest Results.

194 resolved Polymarket markets replayed through the live signal engine. Real resolution data. No curve fitting. The same thresholds running in production.

0%
Win Rate
0
Total Markets
0
Wins
0
Losses
+EV
Expected Value
Politics76%
Geopolitical74%
Economics72%
Crypto70%
Tech68%
Sports64%
Entertainment61%
Theoretical Compounding · $100 Starting Capital
Month Conservative (+10%/mo)Cons. Target (+20%/mo)Target Stretch (+30%/mo)Stretch
Month 1$110$120$130
Month 3$133$173$220
Month 6$177$299$485
Month 12$314$892$2,353

Theoretical projections based solely on 194-market backtest assumptions. Conservative scenario includes daily loss limit triggered 3×/month. Figures do not account for market regime changes, liquidity shifts, or execution slippage. Not financial advice — prediction markets carry risk of total loss.

Signal Intelligence

MANTIS Online.

Running 24/7 on EU infrastructure. Every 5 minutes, active markets are evaluated through the AI pipeline. Positions held until signal degrades or target hits. No manual intervention required.

mantis — signal dashboard — EU-WEST-1
LIVE
● SIGNAL ENGINE: ACTIVE ● RISK GATE: ARMED UPTIME: 127h 43m --:--:-- UTC
MARKET SCAN LOG
    SCANNING next scan: 47s
    OPEN POSITIONS (3)
    UKRAINE-CEASEFIRE-Q2-2026
    YES $8.40 @ 0.31 +$2.10 (+25.0%)
    FED-RATE-CUT-JUNE-2026
    YES $6.20 @ 0.44 +$1.30 (+20.9%)
    EU-GAS-BAN-LIFTED-Q2-2026
    YES $5.00 @ 0.22 -$0.45 (-9.0%)
    TODAY P&L
    +$2.95
    EXPOSURE
    $19.60
    RISK STATUS
    NOMINAL
    EXPOSURE $19.60 / $40.00 49%
    mantis v2.3.0 · rust/tokio · polymarket clob · eu-west-1 system nominal
    Active Strategies

    Four Live Strategies.

    No guessing. Every trade passes a deterministic checklist. MANTIS stays in cash when conditions aren't right — idle hours are never wasted capital.

    maker.rs — Market Making
    // Strategy 1 — Passive market making
    // Places bid/ask quotes on qualifying markets
    // Captures spread as maker rebate

    selection = reward_rate / volume_score
    inventory = Avellaneda-Stoikov skew
    resolution = spread widens near close

    // Strategy 1B — First-Mover
    // Quotes new markets within 500ms of open
    // Captures early-liquidity spread premium
    taker.rs — AI & Arb Execution
    // Strategy 2 — Triple-AI Taker
    // AI consensus scan, high-conviction only
    // Zero-fee markets prioritised

    // Strategy 3 — YES+NO Internal Arb
    // Scans for combined price inefficiencies
    // Both legs fired simultaneously (FOK)

    // Strategy 4 — BTC TA Taker
    // Near-certain markets only (T-10s window)
    // Technical confirmation before entry
    Capital Protection

    8 Hard Risk Rules.

    Every order passes through risk::check() before any execution. A single violation kills the order. No workarounds. No exceptions.

    01
    Exposure Cap
    Total open positions must not exceed $40 including the new bet. Any order that would breach this is rejected.
    Hard cap: $40.00
    02
    Bet Size Cap
    Standard bets truncated to $10 max. Oversize is silently corrected before placement — never rejected, never crashes.
    Max: $10 standard
    03
    Daily Loss Limit
    If today's realised P&L drops below -$15, ALL orders halt for the rest of the UTC day. Logged as DAILY_LOSS_LIMIT_HIT.
    Limit: -$15.00 / day
    04
    Liquidity Filter
    Standard markets must show >$1,000 total volume. Thin markets have wide spreads and poor price discovery — not worth the risk.
    Min volume: $1,000
    05
    Minimum Edge
    Consensus score must clear the conviction threshold. Below this threshold the signal is uncertain — MANTIS stays in cash.
    Minimum conviction threshold
    06
    No Duplicate Positions
    One position per condition_id. Never average down. Never add to a winner. A matching open position returns RiskError::DuplicatePosition.
    1 position per market
    07
    Direction Consensus
    If Groq and Claude disagree on the direction of the trade, the order is rejected entirely. Both models must agree — no split-signal trades reach execution.
    Reject — no consensus
    08
    Maker Position Cap
    Maker orders are hard-capped at $5 per side. Any maker order exceeding this is silently truncated before placement — never rejected, never crashes the engine.
    Max: $5 per side
    System Design

    Rust Architecture Overview.

    Tokio async runtime. Zero-copy channels. DashMap shared state. Sub-50ms order placement. Built to run unattended 24/7.

    Data Feeds
    Gamma API
    Market discovery · 60s poll
    CLOB WebSocket
    Orderbook · price_change · trades
    Price Feed WS
    BTC/USD · real-time tick
    On-Chain
    Polygon RPC · whale tracker
    News Feed
    Tier-1 · Tier-2 · 60–90s poll
    Signal Engine
    Market State
    Shared state · zero-lock reads
    Pre-Screen
    Fast-pass · noise filter
    Oracle
    Deep verify · primary weight
    Consensus
    Weighted score · direction
    Execution
    Risk Gate
    8 hard rules · fail-closed
    Kelly Sizer
    Category mult · max 15%
    CLOB Executor
    Place · cancel · cancel-all
    Exit Monitor
    Trailing stop · rescore 5min
    TUI Dashboard
    terminal UI · positions · P&L
    Progress

    Development Roadmap.

    Phase 1–4
    Build & Validate
    Scaffold, feeds, triple-AI pipeline, risk gate, 194-market backtest — all complete
    Current
    Capital Deploy
    Fund wallet with $100 USDC · run CLOB approvals · enable live trading at conservative limits
    Q2 2026
    Scale Capital
    Increase bet sizes after 4-week live track record · add maker strategy · ARB against Kalshi
    Q3 2026
    Multi-Account
    Separate wallets per strategy · copy-trading signals · institutional position sizing
    Get In Touch

    Questions about MANTIS?

    Built by Orthonode Infrastructure Labs. Research, collaboration, and partnership enquiries welcome.

    Email
    [email protected]
    GitHub
    github.com/orthonode
    Website
    orthonode.xyz
    Twitter / X
    @OrthonodeSys