Offline Play Guide

Run Ollama models headlessly against built-in AI opponents, produce trace files, and upload results to the Global Leaderboard.

Prerequisites

Both files live in the same directory as the main app. Download them to a local folder:

play_offline.js play_offline.sh

Pull a model with Ollama:
```
ollama pull llama3.1:8b
```
Start Ollama with cross-origin support:
```
OLLAMA_ORIGINS=* ollama serve
```
Run a game:
```
node play_offline.js --model llama3.1:8b --opponent lightRush
```
This plays a full game and writes a trace JSON file to the current directory.
Upload to the Global Leaderboard:
```
node play_offline.js --model llama3.1:8b --opponent lightRush --upload
```
Add the --upload flag to automatically submit the result to Firebase when the game finishes.

Flag	Description	Default
`--model <name>`	Ollama model name	llama3.1:8b
`--opponent <ai>`	Built-in AI opponent: random, workerRush, lightRush, heavyRush, economyBoom, turtle, balanced	lightRush
`--max-turns <n>`	Maximum game turns	500
`--interval <n>`	LLM consultation interval in ticks	20
`--ollama-url <url>`	Ollama base URL	http://localhost:11434
`--summoner <name>`	Player name for trace metadata	OfflinePlayer
`--output <file>`	Output trace JSON file path	rts-trace-<timestamp>.json
`--upload`	Upload game result to Firebase after completion	off

After a game finishes, a trace JSON file is saved to your current directory. To replay it:

Try different opponents to see how your model adapts. lightRush is fast and aggressive; turtle is defensive; balanced is a good all-rounder.

Lower --interval (e.g., 10) for more frequent LLM decisions; raise it (e.g., 40) for faster games with fewer API calls.

Use --summoner YourName to tag your name on the leaderboard when uploading results.

Run multiple games with play_offline.sh — it wraps the Node script and can be customized for batch runs.