docs: add CLAUDE.md

Author: S1M0N38

CLAUDE.md

@@ -0,0 +1,71 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+BalatroBench is a static web application that displays performance leaderboards for LLMs playing the card game Balatro. It's a frontend-only project without build tools - the site uses vanilla JavaScript with Tailwind CSS loaded from CDN.
+
+## Architecture
+
+### Core Components
+
+- **index.html**: Main leaderboard page with responsive table layout using Tailwind CSS
+- **script.js**: Fetches and renders leaderboard data from JSON files in the data directory
+- **data/**: Contains benchmark results organized by version and strategy
+  - `data/benchmarks/v0.8.0/default/leaderboard.json`: Primary leaderboard data
+  - Individual model result files in vendor subdirectories (e.g., `openai/gpt-oss-120b.json`)
+
+### Data Structure
+
+The leaderboard displays AI model performance with metrics including:
+- Final round reached (with standard deviation)
+- Success/failure/error rates for API calls
+- Token usage (input/output)
+- Execution time and cost per game
+- Multiple provider usage statistics
+
+Models are identified by `vendor/model` format and ranked by performance metrics.
+
+## Development Commands
+
+### Local Development
+
+```bash
+# Serve the application locally (Python 3)
+python3 -m http.server 8000
+
+# Then visit http://localhost:8000
+```
+
+### File Structure Conventions
+
+- All files use UTF-8 encoding with LF line endings
+- JavaScript: 2-space indentation, 100 character line limit
+- HTML: 2-space indentation, 120 character line limit
+- JSON: 2-space indentation
+
+## Data Management
+
+### Adding New Results
+
+- Leaderboard data is loaded from `data/benchmarks/v0.8.0/default/leaderboard.json`
+- Individual model results stored in `data/benchmarks/v0.8.0/default/[vendor]/[model].json`
+- The application automatically parses `vendor/model` from the `config.model` field
+
+### Result Format
+
+Each entry contains:
+- Run statistics (runs, wins, completion rate)
+- Performance averages and standard deviations
+- API call success metrics
+- Provider usage breakdown
+- Token usage and cost analysis
+
+## Community Contributions
+
+The project accepts AI strategy submissions through a community form process. Contributors develop strategies using the `balatrollm` framework, test locally, then submit via web form for automated server validation and leaderboard inclusion.
+
+## Static Hosting
+
+This is a pure client-side application requiring no backend server - suitable for deployment to static hosting services like GitHub Pages, Netlify, or Vercel.