This repository contains experimental code discussed in the position paper The Modular Imperative: Rethinking LLMs for Maintainable Software by Anastasiya Kravchuk-Kirilyuk, Fernanda Graciolli, and Nada Amin.
The experiments investigate how different development methodologies affect code modularity, maintainability, and architectural quality. The project compares:
- prompt-only or baseline: Direct implementation without architectural guidance from Rubric DSL
- with-rubric: Implementation guided by architectural specifications and constraints via Rubric files
The following table maps the experiment sections from the paper to the corresponding directories in this repository:
| Paper Section | Experiment Description | Repository Location |
|---|---|---|
| 4.1 | Modular Code from Scratch | from-scratch/ |
| 4.2.1 | Porting Code Base to React | porting/ |
| 4.2.2 | Targeted Edits | targeted-edits/ |
| 4.2.2 - Experiment 1 | Simple intra-module edit | targeted-edits/run1/after-exp-1, targeted-edits/run2/after-exp-1 |
| 4.2.2 - Experiment 2 | Complex intra-module edit | targeted-edits/run1/after-exp-2, targeted-edits/run2/after-exp-2 |
| 4.2.2 - Experiment 2a: | Add constraints | targeted-edits/run1/after-exp-2a, targeted-edits/run2/after-exp-2a |
| 4.2.2 - Experiment 3 | Architectural recognition | targeted-edits/run1/after-exp-3, targeted-edits/run2/after-exp-3 |
| 4.2.2 - Experiment 4 | Module integration | targeted-edits/run1/after-exp-4, targeted-edits/run2/after-exp-4 |
Note: Code is in the process of being added to directories. Some directories outlined above may not be complete.
from-scratch/prompt-only/- Implementation without architectural guidancefrom-scratch/with-rubric/- Implementation with Rubric v0.1.0
porting/prompt-only/- Porting existing code without architectural guidanceporting/with-rubric/- Porting with architectural rubrics
targeted-edits/run1/- First experimental runtargeted-edits/run2/- Second experimental run
Note: The targeted-edits experiments used different Rubric versions:
- Run 1: Rubric v0.1.0
- Run 2: Rubric v1.0.0 (includes validation)
MIT