🧠 Agent System Overview

EdgePicks uses a modular, explainable agent architecture to generate informed NFL Pick’em predictions. Each agent contributes specific expertise, enabling diverse and transparent decision-making.

🎯 Agent Roles

Agent	Description
injuryScout	Evaluates how player injuries impact team performance and availability
lineWatcher	Monitors Vegas line movement and betting trends for anomalies
statCruncher	Analyzes season performance data (e.g. yards, penalties, turnovers)
trendsAgent	Detects repeatable historical matchup patterns
guardianAgent	Flags unreliable data, prevents runaway confidence, ensures ethical bounds

🔄 Flow-Orchestrated Predictions

Predictions are orchestrated via flows. Each flow defines:

• Input structure (e.g. home/away team, week)
• Agents to run in sequence or parallel
• How scores and confidence weights are combined
• Streaming logic for real-time feedback via SSE (Server-Sent Events)

📡 Event Streaming Lifecycle

Each agent emits events during a run:

{
  sessionId: string;
  agent: "statCruncher" | "injuryScout" | ...;
  status: "started" | "errored" | "completed";
  confidence: number;
  weightedScore: number;
  durationMs: number;
}
These events are streamed back to the frontend for live rendering and leaderboard updates.

📁 Source Files
lib/flow/runFlow.ts — main flow runner

pages/api/run-agents.ts — streaming prediction endpoint

components/AgentNodeGraph.tsx — real-time agent graph renderer

components/AgentLeaderboardPanel.tsx — ranks agent performance by session