Datasets:
add README landing page
Browse files
README.md
ADDED
|
@@ -0,0 +1,63 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
- ru
|
| 5 |
+
license: mit
|
| 6 |
+
tags:
|
| 7 |
+
- ai-text-detection
|
| 8 |
+
- reproducibility
|
| 9 |
+
- bilingual
|
| 10 |
+
- adversarial-robustness
|
| 11 |
+
- calibration
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# ContentOS — Reproducible Bilingual AI-Text-Detection Ensemble
|
| 15 |
+
|
| 16 |
+
**Pre-print v1.0 (2026-04-27)**
|
| 17 |
+
|
| 18 |
+
This repository contains the open pre-print and supporting artifacts for
|
| 19 |
+
ContentOS, a reproducible English+Russian AI-text-detection ensemble.
|
| 20 |
+
|
| 21 |
+
## Headline numbers (v1.11 production calibration)
|
| 22 |
+
|
| 23 |
+
| Metric | EN | RU |
|
| 24 |
+
|---|---|---|
|
| 25 |
+
| OOD AUROC (44-text smoke) | **0.821** | **0.837** |
|
| 26 |
+
| Wrong-rate | 4% | 9% |
|
| 27 |
+
| p50 latency (EN ensemble) | **1.2 s** | — |
|
| 28 |
+
| Adversarial AUROC (n=300, OOD) | **0.998** | — |
|
| 29 |
+
|
| 30 |
+
## Files
|
| 31 |
+
|
| 32 |
+
- `paper.pdf` — full pre-print (~6,000 words, 9 sections + 5 appendices)
|
| 33 |
+
- `paper.html` — self-contained HTML version with embedded figures
|
| 34 |
+
- `paper.md` — source markdown
|
| 35 |
+
- `figures/` — 4 figures (PNG + SVG)
|
| 36 |
+
- `REPRODUCIBILITY.md` — open methodology, how to reproduce in 90 minutes
|
| 37 |
+
|
| 38 |
+
## Reproducibility
|
| 39 |
+
|
| 40 |
+
Anyone can clone the repo and reproduce all reported numbers:
|
| 41 |
+
|
| 42 |
+
```bash
|
| 43 |
+
git clone github.com/humanswith-ai/greg-personal-claude
|
| 44 |
+
cd greg-personal-claude/services/ml-services-hwai
|
| 45 |
+
pip install -r requirements.txt
|
| 46 |
+
pytest tests/test_calibration_regression.py -v # 8 baselines, 0.05s
|
| 47 |
+
python scripts/eval_ensemble_corpus.py # smoke battery, ~50 min
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
## Cite as
|
| 51 |
+
|
| 52 |
+
```bibtex
|
| 53 |
+
@misc{contentos2026,
|
| 54 |
+
title={ContentOS: A Reproducible Bilingual AI-Text-Detection Ensemble with Adversarial Robustness Evaluation},
|
| 55 |
+
author={Humanswith.ai team},
|
| 56 |
+
year={2026},
|
| 57 |
+
url={https://huggingface.co/datasets/gshevchenko/contentos-preprint},
|
| 58 |
+
}
|
| 59 |
+
```
|
| 60 |
+
|
| 61 |
+
## License
|
| 62 |
+
|
| 63 |
+
MIT for code, methodology, and corpus aggregation. Underlying data sources retain their original licenses (HC3, AINL-Eval-2025, ai-text-detection-pile).
|