JidouChinou Logo
JIDOUCHINOU 自動知能
MULTI-AGENT TRADING DESK
RH-4663 ACTIVE
DOCS (v2.4)
● ROBINHOOD CHAIN (CHAIN ID: 4663): SEQUENCER ONLINE · 118ms FCFS 🛑 ADVERSARIAL VETO ENGINE: 91.4% TOXIC CALLS REJECTED $PONS (pons.family): BONDING CURVE 34.2% · ORGANIC PASS ERC-7579 KERNEL v3.1: SESSION KEYS DELEGATED (-4.5% TRAILING STOP PEG) OffHours BASIS SPREAD: $NVDA TOKEN SPREAD +1.84% vs 4PM NYSE LLM PIPELINE: GROK-4 REASONER UNARY VETO ACTIVE · PESSIMISTIC FAIL-SAFE 🛑 AUTODETECTED HONEYPOT CODES: 1,420 RUGBYTES DROPPED GMGN HUB INTEGRATION: SSE STREAM LIVE (:8800) · 0 LATENCY BUFFER
AUTONOMOUS MULTI-AGENT TRADING DESK ROBINHOOD CHAIN 4663 ERC-7579 SESSION KEYS

A MODEL THAT FAILS
IS A MODEL THAT SAYS NO.

99.4% of AI trading bots go to zero because of the Optimism Trap: single-prompt agents that hallucinate bullish confirmations during toxic pumps and rush to execute buy orders with retail capital.

JidouChinou (自動知能) replaces naive LLM consensus with an institutional 4-Layer · 12-Bot Adversarial Swarm on Robinhood Chain. Governed by formal mathematical invariants, sub-120ms FCFS execution, and a strict Unary Asymmetric Veto: if any invariant fails, capital never moves.

TECHNICAL DOCS FOLLOW ON X
ROBINHOOD CHAIN ECOSYSTEM CONTRACT pons.family FAIR LAUNCH · 100ms FCFS
0x4663...fe6be4fa
BUY ON PONS
BASE GAS: <0.0001 USDC 100% UNARY VETO PROTECTION
VETO RATE
91.4%
TOXIC RUNS DROPPED
FCFS DISPATCH
118ms
SUB-BLOCK LATENCY
SWARM FLEET
12/12
AUTONOMOUS BOTS
FLOW EVALUATED
$48.2M
24H TICK-BY-TICK
JIDOUCHINOU_SWARM // LIVE_INSPECTOR
UNARY VETO: ON
12-Agent Swarm Canvas
● 12-BOT NEURAL TOPOLOGY RUNNING
ACTIVE PIPELINE TRACE TICK #910,482
[L1_PONS_SCOUT] Ingested TokenCreated 0x8a1c.. on pons.family (+92s window)
[L1_BYTECODE] Opcode decompilation: No transfer tax, zero mint trap (PASS)
[L2_ALLOCATOR] Kelly criterion committing 4.2% desk margin ($420 USDC)
[L3_ADVERSARIAL_VETO] Bot #11 PonsChecker: Deployer funded by Mixer (T-14d) -> VETO!
[L4_RISK_MANAGER] Order aborted in 14ms. Zero capital compromised.
// 01. THE PROBLEM

THE OPTIMISM TRAP: WHY 99.4% OF AI TRADING BOTS FAIL

The fundamental flaw of modern AI trading agents is structural sycophancy. Large language models are trained to be agreeable and helpful assistants. When placed in front of a market chart, an agreeable assistant will always find reasons to justify buying.

NAIVE SINGLE-AGENT BOT 99.4% CAPITAL EXTINCTION

Linear Prompting & Sychophantic Hallucination

A typical crypto AI agent operates as a single prompt loop: receives price data -> asks LLM "Should we buy?" -> LLM identifies positive narrative adjectives -> generates a buy order. When a dev launches a copycat honeypot, the bot hallucinates alpha and buys at the top.

  • ✖ Single point of failure: 1 LLM prompt governs execution
  • ✖ Zero adversarial validation: Nobody tries to poke holes in the idea
  • ✖ Fails open: If API times out, it defaults to retry or optimistic execution
  • ✖ Human-scale latency: 550ms–850ms network loops get crushed on FCFS
  • ✖ Unscoped wallet keys: Full private key exposed in server environment
JIDOUCHINOU 12-BOT DESK PESSIMISTIC FALLBACK

Compartmentalized Adversarial Consensus

Intelligence gathering is strictly quarantined from execution. Layer 1 scouts cannot buy; they only extract signals. Layer 3 contains dedicated adversarial checkers whose sole job is to veto trades. Capital only moves when formal safety invariants evaluate to 100% PASS.

  • ✔ Unary Asymmetric Veto: 1 negative invariant halts the entire desk
  • ✔ Pessimistic Fallback: Any timeout, schema error, or network hiccup = NO
  • ✔ Sub-120ms FCFS routing directly into Arbitrum Orbit sequencer feed
  • ✔ ERC-7579 Kernel v3.1 Session Keys: Hard gas, volume, and slippage caps
  • ✔ Front-Exit Engine: Detects cabal exit signatures and front-runs liquidity pulls
// 02. SYSTEM ARCHITECTURE

4 ARCHITECTURE LAYERS · 12 AUTONOMOUS AGENTS

Every trade passes through 4 sequential, adversarial gates. Click any agent below to inspect its sensory role, technical model, input signals, and veto invariants.

#01 // L1 PURE CODE (0$)

PonsScout

Robinhood Chain 4663 WebSocket EVM event sniffer. Filters 90s volume & buyers.

#02 // L1 GROK FAST

BytecodeAuditor

Decompiles contract bytecode on-chain. Flags honeypot, blacklist, and hidden mints.

#03 // L1 NLP PIPELINE

NarrativeScorer

First and second derivatives of social velocity on Telegram, X, and fomo.family.

#04 // L1 OFFHOURS FEED

StockScreener

Tracks 24/7 tokenized stock spreads on Robinhood Chain vs 4PM NYSE close marks.

#05 // L1 GRAPH ENTROPY

InsiderTracker

Disassembles funding co-ancestry across early buyer wallets to isolate cabal syndicates.

#06 // L2 RPC TELEMETRY

RHChainPulse

Monitors FCFS sequencer queue depth. Suspends trading if block latency > 220ms.

#07 // L2 VOLATILITY MODEL

MarketPulse

Classifies macro regime: RISK-ON expansion vs RISK-OFF capital preservation.

#08 // L2 KELLY SIZING

CapitalAllocator

Dynamic Kelly criterion position sizing. Enforces 70/30 desk split & max daily risk.

#09 // L3 GROK-4 REASONER

PonsChecker

UNARY ADVERSARIAL VETO for memecoins. Searches solely for fatal disqualifications.

#10 // L3 CLAUDE SONNET

StockChecker

UNARY ADVERSARIAL VETO for stock tokens. Halts orders if SEC/NYSE halts occur.

#11 // L4 FCFS ENGINE

FrontExitEngine

Sub-second order dispatch. Monitors dev liquidity pulls and front-exits in block 0.

#12 // L4 KERNEL v3.1

PortfolioRiskManager

Shared kill-switch. Enforces -4.5% trailing stop peg and revokes session keys on breach.

#01 // LAYER 1: SENSORY INTELLIGENCE

Bot #1: PonsScout

Primary Role: Real-time EVM event listener on Robinhood Chain Arbitrum Orbit Sequencer
EXECUTION: ZERO LLM COST (PURE CODE)
LATENCY TARGET: < 15ms

PonsScout operates directly inside the Robinhood Chain node via low-latency WebSocket connection. It listens for the `TokenCreated` event emitted by the `pons.family` bonding curve router. Tokens with fewer than 5 unique organic buyers in the initial 90-second window or tokens whose bonding curve is already greater than 40% filled are discarded immediately without invoking any LLM API tokens.

INPUT SIGNALS & TELEMETRY
  • • `event TokenCreated(address indexed token, address indexed creator)`
  • • Tick-by-tick order flow from GMGN DataHub (:8800)
  • • Unique buyer clustering in 90-second observation block
VETO & DISQUALIFICATION THRESHOLD
  • • Unique buyers < 5 within 90 seconds -> AUTO-DROP
  • • Initial liquidity concentration > 35% in single wallet -> AUTO-DROP
  • • Sequencer delay > 250ms during block assembly -> PAUSE DESK
// 03. INTERACTIVE SIMULATOR

TEST THE ADVERSARIAL VETO IN REAL-TIME

Witness how JidouChinou's 4 layers react to live market anomalies. Choose a test case below to dispatch a simulated order through the 12-bot pipeline.

JIDOUCHINOU_DECISION_ENGINE_STREAMS // EMULATOR
STANDBY
// JidouChinou Adversarial Simulator initialized.
// Click one of the buttons above to test execution flow across all 4 tiers.
// Core Invariant: V(T) = Product(I(Invariant_j == PASS))
[SYSTEM] Connected to Robinhood Chain (Chain ID: 4663) Arbitrum Orbit sequencer.
[SYSTEM] GMGN DataHub (:8800) SSE Ingestion online. 12 bots standing by.
Awaiting execution command... UNARY VETO ENFORCED
// 04. FORMAL SPECIFICATION

THE MATHEMATICAL ADVERSARIAL INVARIANT ENGINE

Trade execution in JidouChinou is not a fuzzy probability score or an agreeable LLM prompt. It is a strict multiplicative consensus over 6 non-negotiable mathematical safety invariants.

CONSENSUS CLEARANCE GATE // MULTIPLICATIVE UNANIMITY
VERDICT: [1.0] ORDERS APPROVED
$$\mathcal{V}(T) = \prod_{j=1}^{6} \mathbb{I}\left(\text{Invariant}_j(T) == \text{PASS}\right)$$
\mathcal{V}(T) Clearance Gate ∈ {0, 1} \prod Multiplicative Consensus m = 6 Non-Negotiable Invariants \mathbb{I}(\cdot) Binary Indicator Function PASS Zero Violation Tolerance
⚡ INTERACTIVE CONSENSUS SIMULATOR — TOGGLE INVARIANTS TO OBSERVE UNARY VETO:
V(T) = 1 × 1 × 1 × 1 × 1 × 1 = 1.0 (APPROVED)
● 100% CONSENSUS SATISFIED — ZERO FAILURES
INVARIANT #1

Bytecode Opcode Purity

CATEGORY: EVM DECOMPILATION & STATIC ANALYSIS
$$\text{Hash}_{\text{sim}}(\text{Bytecode}_T, \; \mathcal{D}_{\text{rug}}) < 0.35$$
BOUND: MAX 35% EXPLOIT MATCH
Bytecode_T Candidate Opcode D_rug 120k Exploit Vector DB Hash_sim Jaccard Opcode Metric

Decompiles contract bytecode on-chain. Blocks hidden transfer fee taxes, arbitrary mint hooks, and blacklist mapping manipulations before capital commitment.

INVARIANT #2

Cabal Shannon Entropy

CATEGORY: GRAPH FORENSICS & CO-ANCESTRY
$$H(W_{\text{top20}}) = -\sum_{i=1}^{k} p(w_i) \log_2 p(w_i) \ge 1.65$$
BOUND: MIN 1.65 BITS ENTROPY
W_top20 Top 20 Holder DAG p(w_i) Funder Origin Probability H Shannon Entropy in Bits

Disassembles funding origin trees of early buyers. If 15 ostensibly distinct wallets all trace back to the same parent funder, entropy collapses and triggers an instant veto.

INVARIANT #3

Sequencer Congestion

CATEGORY: ARBITRUM ORBIT LATENCY HEALTH
$$\Delta t_{\text{FCFS}} = t_{\text{receipt}} - t_{\text{submission}} \le 180\text{ms}$$
BOUND: MAX 180ms SUB-BLOCK
t_sub Order WebSocket Injection t_rec Orbit Sequencer Receipt Δt_FCFS Block Inclusion Delay

Monitors Robinhood Chain FCFS block production. If network inclusion latency exceeds 180ms, all trading freezes to prevent execution slippage.

INVARIANT #4

Zero Dev Pre-Allocation

CATEGORY: PONS.FAMILY BONDING INTEGRITY
$$\text{Supply}_{\text{deployer}} == 0.0\text{\%}$$
BOUND: EXACTLY 0.0% PRE-MINE
Supply_deployer Creator Holding % pons.family Fair Launch Protocol

Enforces 100% fair launch on pons.family bonding curves. Any developer pre-allocation or sniper buy in block 0 causes an immediate desk rejection.

INVARIANT #5

Damped Kelly Allocation

CATEGORY: ASYMMETRIC POSITION SIZING
$$f^* = \max\left(0, \; \gamma \cdot \frac{p \cdot b - q}{b} \cdot e^{-\kappa \cdot \sigma}\right)$$
BOUND: FRACTIONAL KELLY γ = 0.35
f* Allocated Capital Fraction p, q Win / Loss Probabilities b Payoff Ratio σ Regime Volatility

Sizes trades mathematically according to edge. Volatility damper e^(-κσ) dynamically reduces position sizing during market turbulence, capping any single trade at 15% of desk margin.

INVARIANT #6

Hard Trailing Stop Peg

CATEGORY: REAL-TIME CAPITAL PRESERVATION
$$\text{StopPrice}(t) = \max_{s \le t} P(s) \cdot (1 - 0.045)$$
BOUND: TIGHT -4.5% TRAILING STOP
P(s) Mark-to-Market Price max P(s) Historical Peak Price 0.955 -4.5% Hard Stop Peg

Continuously ratchets stop-loss upward with price. If price pulls back 4.5% from its highest registered peak, FrontExitEngine injects an immediate sell swap into the FCFS sequencer.

// 05. ARBITRUM ORBIT SEQUENCER

DEFEATING THE 100MS FCFS SEQUENCER MEATGRINDER

Robinhood Chain (Chain ID: 4663) does not have a public mempool or priority gas auctions. The sequencer executes transactions on a strict First-Come, First-Served basis.

HUMAN TRADING WITH SMARTPHONE THUMB 550ms – 850ms (3–8 BLOCKS TOO LATE)
Motor Nerve Conduction (450ms) + Mobile WebSocket Lag (250ms) = 700ms
JIDOUCHINOU MULTI-AGENT SESSION KEY PIPELINE 118ms (BLOCK 0 INCLUSION)
118ms FCFS
1. ZERO MEMPOOL SNOOPING:
Nobody can sandwich your order because Robinhood Chain orders are sequenced privately by the Orbit sequencer.
2. NO PRIORITY GAS WARS:
Bribing miners with high gas does not work. Order execution is determined 100% by latency to the FCFS node.
3. ERC-7579 DELEGATION:
Session keys eliminate pop-up wallet confirmations. Orders dispatch straight to block assembly.
// 06. TRANSPARENCY

WHAT JIDOUCHINOU DOES AND CANNOT DO

We refuse to engage in the deceptive marketing common in crypto AI projects. Here are the unvarnished engineering boundaries of our system.

WHAT THE SYSTEM DOES:

  • ✔ Eliminates 100% of -EV honeypot traps: Bytecode is disassembled before order dispatch, preventing buying tokens with hidden mint traps or frozen transfer hooks.
  • ✔ Protects capital by saying NO: Rejects 91.4% of all launched tokens so you only deploy USDC when mathematical odds favor survival.
  • ✔ Levels the 100ms FCFS playing field: Uses pre-signed ERC-7579 session keys to beat any manual human trader by 400ms to 700ms.
  • ✔ Executes 24/7 OffHours Arbitrage: Capitalizes on tokenized equity spreads when traditional NYSE markets are closed.

WHAT THE SYSTEM CANNOT DO:

  • ✖ It will miss 50x moonshots: If an insider syndicate launches an opaque token with suspicious funding links and pumps it to $20M through wash trading, JidouChinou vetoes it at block 1. We would rather miss 10 pumps than get rugged once.
  • ✖ It cannot predict macro black-swans: Sudden off-chain regulatory interventions or circuit breaker halts cannot be forecasted by on-chain models.
  • ✖ It does not promise infinite returns: Trading is a game of probabilistic survival, not guaranteed wealth. Disciplined risk management remains mandatory.
// 07. FAQ

FREQUENTLY ASKED QUESTIONS

Common architectural questions regarding JidouChinou's multi-agent desk.

How does JidouChinou prevent LLM hallucination during trading? +
We enforce two ironclad rules: (1) Separation of Concerns: LLMs are not allowed to touch execution directly. Sensory and screening bots use pure deterministic code (0$ API cost). (2) Pessimistic Fallback: Adversarial checkers in Layer 3 are rewarded exclusively for vetoing trades. If an LLM response is malformed, times out, or misses a schema field, the function returns approve: false. The system never fails open.
Why is Robinhood Chain (Chain ID: 4663) the primary target? +
Robinhood Chain is an Arbitrum Orbit L2/L3 EVM rollup featuring a private First-Come, First-Served (FCFS) sequencer with 100ms–250ms sub-block confirmation times. Because there is no public mempool, predatory sandwich MEV is eliminated. Transactions compete purely on network latency and automated speed, making it the ideal environment for session-key autonomous desks.
What are ERC-7579 Session Keys and are they safe? +
ERC-7579 is the modular smart account standard (Kernel v3.1). Instead of exposing your main master private key to the trading bot server, you delegate an ephemeral Session Key with cryptographically enforced bounds: maximum daily spend (e.g. 500 USDC), whitelisted router contracts (pons.family only), max slippage (2%), and expiration time (e.g. 7 days). Even if the bot server is completely compromised, attacker funds are restricted by the smart contract.
What happens if GmgnDataHub or RPC goes offline? +
Under the Pessimistic Fallback doctrine, if GmgnDataHub (:8800) or the Robinhood Chain RPC WebSocket disconnects, Bot #12 PortfolioRiskManager immediately enters SAFE_HOLD mode. All pending buy approvals are revoked, existing positions hold with hard on-chain stop-loss pegs, and no new capital is allocated until telemetry confirms 3 consecutive stable blocks.
What are the operating costs and Unit Economics? +
Layer 1 filtering runs on pure Python code ($0). Only filtered candidates (approx. 5% of tokens) are passed to Grok Fast and Grok-4 reasoning. The total estimated API cost is ~$7.20/day (~$216/month). On a $2,000 desk bankroll, break-even requires a +10.8% net monthly return. On a $10,000 bankroll, API overhead drops to just ~2.1%, delivering robust institutional margins.