A ArkenLabs
Technology Behind It

How PRISM Is Built

PRISM combines measured, hybrid, and proxy factor signals into confidence-aware multi-factor snapshots. The public site remains read-only and only serves approved production exports.

1. Data Ingestion & Caching

Waterfalling Providers

Each provider pulls features from public/free sources with explicit waterfalling (e.g., FMP → Finnhub → YFinance). Redundant API calls are avoided using deterministic Evidence Fingerprint Caching.

2. Dual LLM Engine

Fast Scraping & Deep Reasoning

Unstructured data is processed via a Dual LLM architecture: fast local models (Qwen) for unbounded scanning, and high-tier models (GPT-4) for deep, complex policy reasoning.

3. V2 Component Scoring

Structured 0-100 Signals

V2 components emit strict contracts (normalized score, confidence, source type). This keeps measured vs proxy semantics fully transparent.

4. Aggregation & Backtesting

Institutional Grade PIT

Outputs are aggregated via confidence-aware weighting and can be evaluated using our V2 Point-In-Time Backtester for strict leakage prevention.

System Flow

End-to-End Architecture

flowchart TD subgraph Data["External Data Sources"] FMP["Market & Financials"] Gov["Macro & Federal Data"] News["Global News API"] end subgraph Ingestion["Ingestion & Caching Layer"] RL["Rate Limiter & Circuit Breakers"] Cache[("Evidence Cache")] Orch["Parallel Job Orchestrator"] end subgraph LLM["Dual LLM Engine"] Qwen["Fast Local Qwen"] GPT["High-Tier GPT-4"] end subgraph Core["V2 Scoring Components"] SM["Smart Money V2"] SI["State Influence V2"] Lobby["Lobbying V2"] CapFlow["Capital Flow V2"] Fin["Financial Quality V2"] end subgraph Aggregation["Aggregation & Backtesting Layer"] Agg["Confidence-Aware Aggregator"] PIT["Point-In-Time Backtester V2"] DB[("Time-Series DB")] end subgraph Outputs["Public Ecosystem"] API["JSON Exporter"] UI["Static Read-Only UI"] Social["Automated Social Publisher"] end FMP --> RL Gov --> RL News --> RL RL --> Orch Orch <--> Cache Orch --> Qwen Qwen --> GPT Orch --> Fin Orch --> SM Orch --> CapFlow GPT --> SI GPT --> Lobby Fin --> Agg SM --> Agg CapFlow --> Agg SI --> Agg Lobby --> Agg Agg --> DB DB --> PIT Agg --> API PIT --> API API --> UI API --> Social
Run Streams

Daily Snapshot vs Monthly Production

PRISM now supports cadence-aware public/research browsing:

  • Daily production: operational production bootstrap snapshots for high-frequency monitoring.
  • Monthly production: lower-frequency production snapshots for portfolio-level review cadence.
  • All production: complete production history stream when you want no cadence filter.

Cadence selection affects which approved production snapshots appear in rankings and research screens.

Transparency

What Stays Visible

  • Per-component score and confidence summaries.
  • Source-type semantics (`measured`, `hybrid`, `proxy`).
  • Fallback-aware quality controls for public eligibility.
  • Read-only public artifacts generated from approved production outputs.