DUCKWORTH DUGOUT

IPL monitor. Standings, pulse, cap races, match intel. One screen.

Not a cricket score app. An analyst's war room: dark, dense, built for situational awareness. Glance at the Dugout and know who's winning, who's surging, what's at stake tonight, and what the numbers say about tomorrow.

The AI layer synthesizes signals humans would miss. Wire cards surface emerging patterns. Briefings combine H2H records, venue history, and squad news into pre-match tactical reads. Narratives track each franchise's season arc.

The War Room

• Standings · Points table with W/L/NRR and playoff qualification line
• Season Pulse · River chart tracing every franchise's rank trajectory across the season
• Match Timeline · Completed, live, and upcoming fixtures with scores and hero performances
• Cap Race · Orange Cap, Purple Cap, strike rate, economy, and MVP leaderboards
• AI Wire · LLM-generated editorial signals: emerging patterns, records watch, severity-tagged
• Briefing & Dossier · Pre-match tactical read with H2H records, venue history, opposition threat profiles
• Narratives · Per-franchise season arc covering mood, form, momentum, and storylines
• Intel Feed · Aggregated cricket news from ESPNcricinfo, Wisden, CricketAddictor, CricTracker
• Ticker · Scrolling data-driven highlights with career stats, matchup history, and impact context

Architecture

RSS Feeds (Wisden, ESPNcricinfo, CricTracker, CricketAddictor)
  + Wikipedia (fixture scraping, fallback standings)
  + Cricsheet DuckDB (ball-by-ball, career stats)
      |
  SyncContext (shared feeds, standings, today's matches)
      |
  Panel sync modules (ordered, dependency-aware)
      |
  LLM Intelligence (Gemini flash/pro, structured output, cached)
      |
  Static JSON --> frontend/public/api/
      |
  Astro static site --> GitHub Pages

Two-database pattern: Cricsheet data lives in cricket.duckdb (ball-by-ball match data, read-only). The pipeline writes enrichment data (article store, snapshots) to enrichment.duckdb. Cross-DB JOINs bridge the two via DuckDB's ATTACH.

Getting Started

Prerequisites: Python 3.13+ with uv, Node.js 22+, Gemini API key (optional, data panels work without it)

git clone https://github.com/ankitksr/duckworth-dugout.git
cd duckworth-dugout

# Pipeline
cp .env.example .env              # set CT_LLM_API_KEY if you have one
uv sync
uv run python -m pipeline sync    # fetch data + run intelligence
# or: uv run python -m pipeline seed-sample   # use sample JSON for dev

# Frontend
cd frontend && npm install
npm run dev                        # localhost:4321

[!NOTE] LLM-powered panels (wire, briefing, dossier, narratives, scenarios) require a Gemini API key. All data panels (standings, schedule, caps, pulse) work without one.

Pipeline

Panels are organized into tiers by refresh frequency:

Tier	Panels	Interval
hot	intel_log, wire	~5 min
warm	standings, caps, schedule, ticker, pulse	Periodic
cool	scenarios, records, briefing, narratives, dossier, match_notes	On demand / hash-triggered

uv run python -m pipeline sync                          # all tiers
uv run python -m pipeline sync --tiers hot              # hot tier only
uv run python -m pipeline sync --tiers hot,warm         # multiple tiers
uv run python -m pipeline sync --panel standings        # single panel
uv run python -m pipeline sync --watch                  # continuous (5-min default)
uv run python -m pipeline sync --watch --interval 120   # custom interval (seconds)
uv run python -m pipeline sync --force                  # bypass caches

Configuration

All config via env vars (see .env.example):

Variable	Required	Description
CT_LLM_API_KEY	Yes (for LLM panels)	Gemini API key
CT_LLM_VERTEX	No	Use Vertex AI (true/false)
CT_LLM_GCP_PROJECT	If Vertex	GCP project ID
CT_LLM_GCP_LOCATION	If Vertex	GCP region (default: us-central1)
CT_LLM_MODEL	No	Flash model override (default: gemini-2.5-flash)
CT_LLM_MODEL_PRO	No	Pro model override (default: gemini-3-pro-preview)
CT_LLM_RATE_LIMIT_RPM	No	Rate limit (default: 10 req/min)
CRICKET_DB_PATH	No	Cricsheet DuckDB path (default: data/cricket.duckdb)

Project Structure

duckworth-dugout/
├── frontend/              # Astro + React static site
│   ├── src/
│   │   ├── components/    # React components (WarRoomView, panels, bridge)
│   │   ├── hooks/         # Data fetching & state management
│   │   ├── types/         # TypeScript interfaces
│   │   ├── styles/        # CSS (dark theme, grid layout)
│   │   └── pages/         # Astro entry point
│   └── public/api/        # Static JSON (pipeline output)
├── pipeline/              # Python data pipeline
│   ├── panels/            # Per-panel sync modules
│   ├── sources/           # Data source parsers (RSS, Wikipedia, Cricsheet)
│   ├── intel/             # LLM intelligence generators + prompt templates
│   ├── llm/               # LLM provider abstraction (Gemini)
│   ├── db/                # DuckDB schema & connection
│   └── ipl/               # Franchise metadata
├── data/                  # Pipeline output (JSON + DuckDB)
└── docs/                  # Screenshots

_{MIT License · Built by ankitksr}