idea_world_labDEV JOURNAL

Idea World Lab

A personal lab experimenting and documenting Godot, data collection, and tool creation to build idea_world.

Repository Direction

It started as a repository for making games. I wanted to implement scenes I wished to experience in real life, like a farming simulation, inside a game.

Now the scope has broadened to better understand Godot, and to create tools that explain, fix, and transform Godot code. Ultimately, I want to build the idea_world I envision using those experiences and tools.

Therefore this repository is closer to an idea_world_lab. It is not just a place to let ideas drift by, but a workspace where I record experiments, retrospectives, architecture, data collection, model‑training preparation, and game‑development directions needed to create idea_world.

It is not meant to look like a finished product. It includes failed experiments and shaky judgments, documenting how I thought and iterated.

Project Nature

This project does not want to end as a simple RAG example that merely copies Godot documentation. It records a production flow that gathers material based on the official docs, preserves the original text, separates table roles, matches sources, attaches validators, and uses a debugger to confirm failures.

The Godot official documentation and external assets inside the repository follow their original licenses and are noted separately. All other retrospectives, diagrams, screenshots, validation flows, explanatory order, and tool‑design records are original work.

The project does not aim to monopolize the abstract RAG concept itself. It seeks to learn and share within the open‑source ecosystem while clearly preserving the sources and context of my own failure records and design flows.

Licenses and external source references are organized separately in LICENSE and THIRD_PARTY_NOTICES.md.

The Game I Originally Wanted to Make

The first game that came to mind was a farming simulation. I wanted to capture the calm I felt in Stardew Valley, the sensation of living quietly on a farm, and implement a life that is hard to pursue in reality inside a game.

Now I am first building Godot models and tools that will help make that game better, rather than focusing on game production itself. The direction hasn’t disappeared; it has expanded into foundational work for the game.

Current Interests

  • Collecting Godot official documentation and converting to JSONL
  • Organizing evidence for Godot 3 → Godot 4 migration
  • Designing datasets for Godot code explanation / transformation
  • Experimenting with Retriever, Validator, LLM call flows
  • Creating tools and models that can later be linked back to game production

Development / Automation Flow

  • LLM: Experiments based on Qwen 3.6 35B. Validating Godot doc classification, code explanation, and migration data generation flows
  • AI PR Review: Separate from the CodeRabbit GitHub App, importing the qwen‑code workflow and customizing it to my taste for PR comments
  • AI PR Review API: Initially stopped due to cost, now retrying by switching to the OpenRouter free tier using openai/gpt‑oss‑120b:free
  • Static CI: GitHub Actions + ESLint
  • Code Score: Automatic evaluation with CodeFactor A–F
  • Assets: Reviewing free/CC0 assets such as Sprout Lands, FarmLands, SpriteCook, Kenney, etc.