🎲 v1.0.0 — The Public Release
🎲 Likelihoodlum v1.0.0
First public release. Scores GitHub repos 0–100 for LLM-generated code using nine heuristic signals.
Highlights
- Nine scoring signals: code velocity, session productivity, commit size uniformity, message patterns, burst detection, multi-author discount, extreme per-commit velocity, project-scale plausibility, and generated file ratio
- Bidirectional scoring: suspicious patterns push the score up, clearly human patterns pull it down
- Zero dependencies: stdlib-only Python 3.10+
- Concurrent fetching: up to 10 parallel API requests for fast analysis
- Bot filtering: dependabot and similar bots excluded from analysis and API calls
- Project-scale plausibility: compares total output against the repo's true creation date — shipping 60K lines in a day gets flagged
- pip installable:
pip install git+https://github.com/gotnull/likelihoodlum.git→likelihoodlum owner/repo
Wall of Truth Highlights
| Repository | Score | Verdict |
|---|---|---|
| anthropics/claudes-c-compiler | 81/100 | 🤖 Very likely LLM-generated |
| openai/codex | 35/100 | 🤔 Possibly LLM-assisted |
| facebook/react | 10/100 | 👤 Almost certainly human-written |
| django/django | 0/100 | 👤 Almost certainly human-written |
| bitcoin/bitcoin | 0/100 | 👤 Almost certainly human-written |
| golang/go | 0/100 | 👤 Almost certainly human-written |
Install
pip install git+https://github.com/gotnull/likelihoodlum.git
likelihoodlum owner/repoSee the full changelog for details.