Agencio Predict - All-Seeing Eye Architecture

System Overview

The All-Seeing Eye is the central nervous system of Agencio Predict, aggregating data from 15+ source categories into a unified intelligence layer that powers insights, alerts, and trading decisions.

Data Source Categories

Scheduler Jobs

5min

Aggregation Cycle

436

DSL Primitives

Core Capabilities

1
Data Aggregation — Parallel fetch from market data, prediction markets, sentiment, derivatives, institutional holdings, and more
2
Correlation Analysis — Cross-source correlation matrix tracking regime changes and anomalies
3
Insight Generation — AI-powered insight synthesis with confidence scoring
4
Black Swan Detection — Real-time anomaly detection with historical pattern matching
5
Action Generation — Automated trading actions with guardrail checks

Data Flow Pipeline

The All-Seeing Eye orchestrator runs on a 5-minute cycle, aggregating all data sources and generating insights.

flowchart TB subgraph Sources["Data Sources (15 Categories)"] M[Market Data
Yahoo, CoinGecko, Finnhub] P[Prediction Markets
Polymarket, Kalshi, Metaculus] S[Sentiment
Reddit, Twitter, News] D[Derivatives
VIX, Funding, OI] I[Institutional
SEC 13F Holdings] A[Activist
SEC 13D/13G] O[Options Flow
Polygon.io] W[Wallet Clustering
Polymarket Fraud] R[Resilience
Crisis Scoring] SC[Supply Chain
Company Relations] PA[Price Action
Patterns, FVG] AL[Algorithms
Performance] AI[AI Predictions
Internal Models] MK[Marketing
Ad Signals] TR[Trust Layer
Whale/Bot Detection] end subgraph Aggregator["All-Seeing Eye Aggregator"] AGG[aggregateAllData
Parallel Fetch] NORM[Normalize to
UnifiedDataPoint] CORR[Correlation
Matrix] end subgraph Intelligence["Intelligence Layer"] INS[Insight
Generator] BS[Black Swan
Detector] ACT[Action
Generator] end subgraph Consumers["Consumers"] EXEC[Executor
Live Run Kill] DSL[DSL Primitives
Algorithm Builder] UI[Dashboard
Alerts & Insights] SSE[SSE Stream
Real-time Updates] end M --> AGG P --> AGG S --> AGG D --> AGG I --> AGG A --> AGG O --> AGG W --> AGG R --> AGG SC --> AGG PA --> AGG AL --> AGG AI --> AGG MK --> AGG TR --> AGG AGG --> NORM NORM --> CORR CORR --> INS CORR --> BS INS --> ACT BS --> EXEC INS --> DSL ACT --> UI INS --> SSE style AGG fill:#22d3ee,color:#000 style BS fill:#f85149,color:#fff style EXEC fill:#f0883e,color:#000

Data Source Categories

Each source category provides UnifiedDataPoint objects with normalized values, confidence scores, and metadata.

Category	Sources	Refresh Rate	Key Signals
Market	Yahoo, CoinGecko, Finnhub, Frankfurter	15 min	Price, volume, momentum
Prediction	Polymarket, Kalshi, Metaculus, PredictIt	15 min	Event probabilities, market consensus
Sentiment	Reddit, Twitter, Telegram, Discord, News	15 min	Social sentiment, trending topics
Derivatives	CBOE VIX, Fear/Greed, Binance, Coinglass	5 min	VIX, funding rates, open interest
Institutional	SEC EDGAR 13F	Weekly	Ownership %, position changes
Activist	SEC EDGAR 13D/13G	Daily	Activist positions, campaigns
Options Flow	Polygon.io	5 min	Put/call ratio, unusual activity
Wallet Clustering	Polymarket CLOB	6 hours	Fraud detection, manipulation signals
Resilience	Historical crisis data	Weekly	Crisis performance, defensive scores
Supply Chain	FactSet, Finnhub	Weekly	Customer/supplier relationships
Price Action	Internal pattern detection	Real-time	FVG, candlestick patterns, S/R
Algorithms	Internal performance tracking	Per tick	Strategy performance metrics
AI Predictions	PINN, LSTM, XGBoost models	Daily	Price forecasts, regime predictions
Marketing	Google, Meta, TikTok, LinkedIn	Hourly	Ad performance signals
Trust	Whale/bot classification	15 min	VPIN, Kyle's Lambda, bot signatures

Data Aggregation

The aggregateAllData() function performs parallel fetches from all source categories and normalizes data to UnifiedDataPoint format.

flowchart LR subgraph Input["Raw Data"] R1[VIX: 18.5] R2[BTC Funding: 0.01%] R3[Reddit Sentiment: 0.7] R4[P/C Ratio: 1.2] end subgraph Normalize["Normalize to -1 to +1"] N1[VIX: +0.1
Slightly elevated] N2[Funding: -0.2
Bearish] N3[Sentiment: +0.4
Bullish] N4[P/C: +0.2
Put heavy] end subgraph Unified["UnifiedDataPoint"] U[sourceId: string
sourceName: string
sourceType: string
normalizedValue: number
rawValue: number
direction: up/down/flat
momentum: number
confidence: number
qualityScore: number
timestamp: string
metadata: object] end R1 --> N1 R2 --> N2 R3 --> N3 R4 --> N4 N1 --> U N2 --> U N3 --> U N4 --> U

AggregatedData Structure

Field	Type	Description
`market`	UnifiedDataPoint[]	Market data signals
`prediction`	UnifiedDataPoint[]	Prediction market signals
`sentiment`	UnifiedDataPoint[]	Social sentiment signals
`derivatives`	UnifiedDataPoint[]	Derivatives market signals
`institutional`	UnifiedDataPoint[]	13F institutional holdings
`activist`	UnifiedDataPoint[]	13D/13G activist positions
`optionsFlow`	UnifiedDataPoint[]	Options unusual activity
`walletClustering`	UnifiedDataPoint[]	Polymarket fraud signals
`resilience`	UnifiedDataPoint[]	Crisis resilience scores
`supplyChain`	UnifiedDataPoint[]	Company relationships
`priceAction`	UnifiedDataPoint[]	Pattern detection
`algorithms`	UnifiedDataPoint[]	Strategy performance
`ai`	UnifiedDataPoint[]	AI model predictions
`marketing`	UnifiedDataPoint[]	Ad platform signals
`trust`	UnifiedDataPoint[]	Whale/bot classification
`all`	UnifiedDataPoint[]	Combined array of all sources
`stats`	object	totalDataPoints, avgConfidence, avgQuality

Executor Integration

The All-Seeing Eye's black swan detection is directly wired to the trading executor. When a high-severity warning is detected, all live runs are automatically halted.

sequenceDiagram participant E as Executor tickRun() participant ASE as All-Seeing Eye participant BS as Black Swan Detector participant DB as Database E->>ASE: listActiveWarnings() ASE->>BS: Check warnings cache BS-->>ASE: Active warnings list alt High+ severity warning exists ASE-->>E: Warning (severity: high/severe/extreme) E->>DB: Record guardrail kill E->>DB: Update run ended_at, kill_reason E->>E: Return killed:L1:black_swan else No severe warnings ASE-->>E: Empty or low-severity warnings E->>E: Continue normal tick processing end

Kill Switch Hierarchy

The executor checks multiple kill conditions before processing each tick:

1
Platform AI Kill Switch — Admin-triggered emergency stop for all AI trading
2
Black Swan Warning — Automatic halt when All-Seeing Eye detects high+ severity warning
3
Broker Credentials — Kill if broker credentials are missing during live run
4
Position Reconciliation — Kill if broker positions don't match internal state
5
L1/L2/L3 Guardrails — Strategy-specific risk limits

Scheduler Jobs

87 scheduler jobs feed data into the All-Seeing Eye. Key jobs related to the intelligence layer:

Job Name	Interval	Description
`all-seeing-eye-orchestrator`	5 min	Main aggregation cycle
`sync-polymarket-bets`	6 hours	Fetch CLOB trades for wallet clustering
`wallet-clustering-incremental`	Daily	Analyze recently active wallets
`sync-13f-incremental`	Weekly	Fetch SEC 13F filings
`sync-activist-filings`	Daily	Fetch SEC 13D/13G filings
`sync-resilience-scores`	Weekly	Compute crisis resilience scores
`human-automation`	15 min	Divergence engine + composite score
`sentiment-rollup`	15 min	Aggregate social sentiment

Key Files

File	Purpose
`packages/be/src/all-seeing-eye/index.ts`	Main module exports
`packages/be/src/all-seeing-eye/orchestrator/index.ts`	Orchestrator (start, stop, runCycle)
`packages/be/src/all-seeing-eye/aggregation/index.ts`	Data aggregation (aggregateAllData)
`packages/be/src/all-seeing-eye/aggregation/options-flow.ts`	Options flow bridge
`packages/be/src/all-seeing-eye/black-swan/detector.ts`	Black swan detection
`packages/be/src/all-seeing-eye/insights/index.ts`	Insight generation
`packages/be/src/all-seeing-eye/actions/index.ts`	Action generation
`packages/be/src/scheduler/index.ts`	Scheduler job registration