TypeScript agent that predicts FIFA World Cup 2026 match outcomes, group standings, and tournament winners using an ensemble of Elo, Poisson, and form-based models — with optional AI hybrid blending and value-bet detection.

• Match predictions — win/draw/loss probabilities and expected goals per fixture
• Group standings — simulated points tables from predicted results
• Tournament simulation — projected champion and semifinalists
• Value bet detection — compare model vs market odds with Kelly sizing
• Agent tools — extensible tool-based prediction orchestration
• AI hybrid mode — blend statistical ensemble with LLM qualitative analysis
• Redis cache — optional caching for predictions and comparisons

Requirements: Node.js 20+

npm install
npm test
npm run predict -- configs
npm run predict -- predict A1
npm run predict -- clubs Arsenal Chelsea
npm run predict -- standings
npm run predict -- value-bets

npm run predict -- clubs Arsenal Chelsea
npm run predict -- clubs Arsenal Chelsea --mock-ai
npm run predict -- predict A1
npm run predict -- predict A1 --config elo-heavy
npm run predict -- predict A1 --ai
npm run predict -- hybrid A1 --ai-weight 0.4
npm run predict -- compare A1
npm run predict -- tournament --config host-bias
npm run predict -- value-bets

Predict domestic club fixtures by team name — separate from the World Cup matchId flow and from sports-betting-toolbox. Uses PredictionAgent.predictClubs() and the predict_clubs agent tool.

1. Loads remote soccer CSV history (England, Spain, Italy, Germany, France)
2. Builds team form and head-to-head context
3. Blends statistical baseline with AI via the existing ai/ provider pattern

cp .env.example .env   # set OPENAI_API_KEY for live AI
npm run predict -- clubs Arsenal Chelsea
npm run predict -- clubs Arsenal Chelsea --mock-ai   # offline
npm run predict -- clubs Arsenal Chelsea --no-ai      # statistical only

Output: home/draw/away probabilities, predicted outcome, confidence, and AI reasoning.

Combine statistical models with an LLM for qualitative factors (form, tactics, host pressure).

Copy .env.example to .env and set your key. Works with any OpenAI-compatible endpoint via AI_BASE_URL.

AI_BLEND_WEIGHT=0.3 npm run predict -- hybrid A1

Pass --config <name> to any command, or use compare to see all configs side-by-side.

fifa-world-cup-2026-prediction-agent/
├── src/
│   ├── agent/          Prediction agent and tools
│   ├── ai/             AI provider, prompt builder, hybrid combiner
│   ├── cli/            Command-line interface
│   ├── config/         Prediction + AI config profiles
│   ├── data/           Teams, groups, venues, fixtures, club CSV history
│   ├── models/         Elo, Poisson, form, ensemble
│   ├── odds/           Market odds and value bets
│   ├── predictions/    Match, club, and tournament predictors
│   ├── types/          TypeScript interfaces
│   └── utils/          Kelly, EV, logging, Redis cache
└── tests/

import { predictMatch } from "./src/predictions/matchPredictor.js";
import { findValueBets } from "./src/odds/valueBets.js";

const result = predictMatch("A1", "balanced");
const edges = findValueBets();