README

Language Model Code Complexity metric for PHP codebases.

LM-CC measures code complexity from the LLM's perspective. Based on arXiv:2602.07882, it correlates at r=-0.92 to -0.97 with LLM task performance — lowering LM-CC improves LLM repair (+20.9%) and translation (+10.2%) success rates.

Installation

composer require k-kinzal/lm-cc-php

Requires an LLM backend:

# Option A: Ollama (recommended, free, local)
ollama pull llama3
ollama serve

# Option B: OpenAI API
export LMCC_API_KEY=sk-...

Quick Start

# Analyze source directory
vendor/bin/lm-cc analyze src/

# With CI threshold (exit code 1 if exceeded)
vendor/bin/lm-cc analyze src/ --threshold 50.0

# JSON output
vendor/bin/lm-cc analyze src/ --format json

# Generate config file
vendor/bin/lm-cc init

Configuration

Place lm-cc.yaml in your project root:

backend: ollama
model: llama3
base_url: "http://localhost:11434"
alpha: 0.8
percentile: 67.0
threshold: 50.0
format: text
exclude:
  - "vendor/*"
  - "tests/*"

See docs/usage.md for all options.

CI Integration

# GitHub Actions
- name: LM-CC Analysis
  run: vendor/bin/lm-cc analyze src/ --threshold 50.0

See docs/ci-integration.md for full examples.

Documentation

• What is LM-CC? — Theory and formula
• Installation — Backend setup
• Usage — CLI reference
• CI Integration — Pipeline setup

Acknowledgments

This tool is an implementation of the LM-CC metric proposed in:

Mingjie Liu, Zichao Li, Chengyu Huang, Yihong Dong, Yongqiang Chen, Shuijing He, Bo Zheng, and Ge Li. LM-CC: Language Model Code Complexity — A New Metric from the LLM's Perspective. arXiv:2602.07882, 2025.

License

MIT — see LICENSE.