lmscan 0.1.0__tar.gz

lmscan-0.1.0/.github/workflows/ci.yml

@@ -0,0 +1,21 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python-version }}
+      - run: pip install -e . pytest
+      - run: python -m pytest tests/ -v --tb=short

lmscan-0.1.0/.gitignore

@@ -0,0 +1,14 @@
+__pycache__/
+*.pyc
+*.pyo
+dist/
+build/
+*.egg-info/
+.eggs/
+*.egg
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+.venv/
+venv/
+.env

lmscan-0.1.0/.pre-commit-hooks.yaml

@@ -0,0 +1,8 @@
+- id: lmscan
+  name: lmscan
+  description: Detect AI-generated text
+  entry: lmscan
+  language: python
+  types: [text]
+  pass_filenames: true
+  args: ["--threshold", "0.7"]

lmscan-0.1.0/CHANGELOG.md

@@ -0,0 +1,16 @@
+# Changelog
+
+## [0.1.0] - 2025-04-10
+
+### Added
+- Statistical AI text detection using 12 linguistic features
+- Model fingerprinting for GPT-4, Claude, Gemini, Llama, Mistral
+- Per-sentence analysis with individual AI probability scores
+- CLI with text/file/stdin input, JSON output, threshold gating
+- Python API: `scan()`, `scan_file()`
+- Burstiness, entropy, Zipf deviation, vocabulary richness analysis
+- AI "slop word" detection (known LLM vocabulary markers)
+- Transition word ratio, readability consistency, bigram/trigram repetition
+- Beautiful ASCII terminal report with feature table and model attribution
+- Zero external dependencies
+- 96 tests

lmscan-0.1.0/LICENSE

@@ -0,0 +1,17 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   Copyright 2025 Zacharie B
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

lmscan-0.1.0/PKG-INFO

@@ -0,0 +1,251 @@
+Metadata-Version: 2.4
+Name: lmscan
+Version: 0.1.0
+Summary: Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline.
+Project-URL: Homepage, https://github.com/stef41/lmscan
+Project-URL: Repository, https://github.com/stef41/lmscan
+Project-URL: Issues, https://github.com/stef41/lmscan/issues
+Project-URL: Changelog, https://github.com/stef41/lmscan/blob/main/CHANGELOG.md
+Author: Zacharie B
+License: Apache-2.0
+License-File: LICENSE
+Keywords: ai-detection,ai-text-detection,chatgpt,claude,content-detection,gemini,gptzero,llm,plagiarism,text-forensics
+Classifier: Development Status :: 4 - Beta
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: Intended Audience :: Education
+Classifier: Intended Audience :: Science/Research
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Text Processing :: Linguistic
+Classifier: Typing :: Typed
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+
+# 🔍 lmscan
+
+**Detect AI-generated text. Fingerprint which LLM wrote it. Open-source GPTZero alternative.**
+
+[![PyPI](https://img.shields.io/pypi/v/lmscan?color=blue)](https://pypi.org/project/lmscan/)
+[![License](https://img.shields.io/badge/license-Apache--2.0-green)](LICENSE)
+[![Python](https://img.shields.io/pypi/pyversions/lmscan)](https://pypi.org/project/lmscan/)
+[![Tests](https://img.shields.io/badge/tests-96%20passed-brightgreen)]()
+
+> GPTZero charges $15/month. Originality.ai charges per scan. Turnitin locks you into institutional contracts.
+>
+> **lmscan is free, open-source, works offline, and tells you _which_ model wrote the text.**
+
+```
+$ lmscan "In today's rapidly evolving digital landscape, it's important
+to note that artificial intelligence has become a pivotal force in
+transforming how we navigate the complexities of modern life..."
+
+🔍 lmscan v0.1.0 — AI Text Forensics
+══════════════════════════════════════════════════
+
+  Verdict:     🤖 Likely AI (77% confidence)
+  Words:       184
+  Sentences:   10
+  Scanned in 0.01s
+
+┌────────────────────────────┬──────────┬────────────────────┐
+│ Feature                    │ Value    │ Signal             │
+├────────────────────────────┼──────────┼────────────────────┤
+│ Burstiness                 │ 0.07     │ 🔴 Very low (AI)    │
+│ Sentence length variance   │ 0.27     │ 🟡 Below average    │
+│ Slop word density          │ 20.7%    │ 🔴 High (AI)        │
+│ Transition word ratio      │ 2.2%     │ 🟡 Elevated         │
+│ Readability consistency    │ 0.00     │ 🔴 Very low (AI)    │
+│ ...                        │          │                     │
+└────────────────────────────┴──────────┴────────────────────┘
+
+🔎 Model Attribution
+  1. GPT-4 / ChatGPT    62% — "delve", "tapestry", "beacon", "landscape" (×2), +19 more
+  2. Claude (Anthropic)  13% — "robust", "nuanced", "comprehensive"
+  3. Gemini (Google)      9% — "furthermore", "additionally"
+
+⚠️  Flags
+  • Very low burstiness (0.07) — AI text is more uniform in complexity
+  • High slop word density (20.7%) — contains known AI vocabulary markers
+```
+
+## Install
+
+```bash
+pip install lmscan
+```
+
+**Zero dependencies.** Works with Python 3.9+. No API keys. No internet. No GPU.
+
+## Usage
+
+```bash
+# Scan text directly
+lmscan "Your text here..."
+
+# Scan a file
+lmscan document.txt
+
+# Pipe from stdin
+cat essay.txt | lmscan -
+
+# JSON output (for scripts and CI)
+lmscan document.txt --format json
+
+# Per-sentence breakdown
+lmscan document.txt --sentences
+
+# CI gate: fail if AI probability > 50%
+lmscan submission.txt --threshold 0.5
+```
+
+### Python API
+
+```python
+from lmscan import scan
+
+result = scan("Text to analyze...")
+
+print(f"AI probability: {result.ai_probability:.0%}")
+print(f"Verdict: {result.verdict}")
+print(f"Confidence: {result.confidence}")
+
+# Which model wrote it?
+for model in result.model_attribution:
+    print(f"  {model.model}: {model.confidence:.0%}")
+    for evidence in model.evidence[:3]:
+        print(f"    → {evidence}")
+
+# Per-sentence analysis
+for sentence in result.sentence_scores:
+    if sentence.ai_probability > 0.7:
+        print(f"  🤖 {sentence.text[:60]}... ({sentence.ai_probability:.0%})")
+```
+
+### Scan entire directories
+
+```python
+from lmscan import scan_file
+import glob
+
+for path in glob.glob("submissions/*.txt"):
+    result = scan_file(path)
+    print(f"{path}: {result.verdict} ({result.ai_probability:.0%})")
+```
+
+## How It Works
+
+lmscan uses **12 statistical features** derived from computational linguistics research to distinguish AI-generated text from human writing:
+
+| Feature | What it measures | AI signal |
+|---------|-----------------|-----------|
+| **Burstiness** | Variance in sentence complexity | AI text is unusually uniform |
+| **Sentence length variance** | How much sentence lengths vary | AI produces uniform lengths |
+| **Vocabulary richness** | Type-token ratio (Yule's K corrected) | AI reuses words more |
+| **Hapax legomena ratio** | Fraction of words appearing once | AI has fewer unique words |
+| **Zipf deviation** | How word frequencies follow Zipf's law | AI deviates from natural distribution |
+| **Readability consistency** | Flesch-Kincaid variance across paragraphs | AI maintains constant readability |
+| **Bigram/trigram repetition** | Repeated word pairs and triples | AI repeats phrase structures |
+| **Transition word ratio** | "however", "moreover", "furthermore"... | AI overuses transitions |
+| **Slop word density** | Known AI vocabulary markers | "delve", "tapestry", "beacon"... |
+| **Punctuation entropy** | Diversity of punctuation usage | AI is more predictable |
+
+Each feature produces a signal via sigmoid transformation. The weighted combination produces the final AI probability.
+
+### Model Fingerprinting
+
+lmscan includes vocabulary fingerprints for 5 major LLM families:
+
+| Model | Distinctive markers |
+|-------|-------------------|
+| **GPT-4 / ChatGPT** | "delve", "tapestry", "landscape", "leverage", "multifaceted", "it's important to note" |
+| **Claude (Anthropic)** | "certainly", "I'd be happy to", "straightforward", "I should note" |
+| **Gemini (Google)** | "crucial", "here's a breakdown", "keep in mind" |
+| **Llama / Meta** | "awesome", "fantastic", "hope this helps" |
+| **Mistral / Mixtral** | "indeed", "moreover", "hence", "noteworthy" |
+
+Attribution uses weighted vocabulary matching, phrase detection, and hedging pattern analysis.
+
+## Accuracy & Limitations
+
+**What lmscan is good at:**
+- Detecting text with strong AI stylistic patterns
+- Identifying which model family generated text
+- Scanning at scale (thousands of documents) with zero cost
+- Providing explainable evidence (not a black box)
+
+**What lmscan cannot do:**
+- Detect AI text that has been manually edited or paraphrased
+- Work reliably on very short text (<50 words)
+- Detect AI text in non-English languages (English-only for now)
+- Replace human judgment — use as a signal, not a verdict
+
+**This is statistical analysis, not a neural classifier.** It detects stylistic patterns, not watermarks. It works best on unedited LLM output and degrades gracefully on edited text.
+
+## CI Integration
+
+### GitHub Actions
+
+```yaml
+- name: AI Content Check
+  run: |
+    pip install lmscan
+    lmscan submission.txt --threshold 0.7 --format json
+```
+
+### Pre-commit
+
+```yaml
+repos:
+  - repo: https://github.com/stef41/lmscan
+    rev: v0.1.0
+    hooks:
+      - id: lmscan
+        args: ["--threshold", "0.7"]
+```
+
+## Research Background
+
+lmscan's approach is informed by published research on AI text detection:
+
+- **DetectGPT** (Mitchell et al., 2023) — perturbation-based detection using log probability curvature
+- **GLTR** (Gehrmann et al., 2019) — statistical visualization of token predictions
+- **Binoculars** (Hans et al., 2024) — cross-model perplexity comparison
+- **Zipf's Law in NLP** — word frequency distributions differ between human and AI text
+- **Stylometry** — decades of authorship attribution research applied to AI forensics
+
+lmscan takes the statistical intuitions from these papers and implements them as lightweight, dependency-free heuristics that work without requiring a reference language model.
+
+## FAQ
+
+**Q: Is this as accurate as GPTZero?**
+A: GPTZero uses neural classifiers trained on labeled data. lmscan uses statistical heuristics. GPTZero is more accurate on edge cases; lmscan is free, offline, and explainable. Use both if accuracy matters.
+
+**Q: Can students use this to evade AI detection?**
+A: lmscan shows which features trigger detection, which could help someone understand why text reads as AI-generated. This is by design — understanding AI writing patterns makes everyone a better writer. The same information is available in published research papers.
+
+**Q: Does it work on non-English text?**
+A: Currently English-only. The slop word lists and transition word lists are English-specific. Statistical features (entropy, burstiness) work across languages but haven't been calibrated.
+
+**Q: Does it phone home?**
+A: No. Zero network requests. No telemetry. No API keys. Everything runs locally.
+
+**Q: How is model attribution possible without running the model?**
+A: Each LLM family has characteristic vocabulary biases. GPT-4 loves "delve" and "tapestry". Claude says "I'd be happy to". These are statistical fingerprints — not guaranteed attribution, but strong signals.
+
+## See Also
+
+- [reverse-SynthID](https://github.com/aloshdenny/reverse-SynthID) — Reverse-engineering Google's image watermarking
+- [vibesafe](https://github.com/stef41/vibesafe) — AI code safety scanner
+- [injectionguard](https://github.com/stef41/injectionguard) — Prompt injection detection
+- [vibescore](https://github.com/stef41/vibescore) — Grade your vibe-coded project
+
+## License
+
+Apache-2.0

lmscan-0.1.0/README.md

@@ -0,0 +1,221 @@
+# 🔍 lmscan
+
+**Detect AI-generated text. Fingerprint which LLM wrote it. Open-source GPTZero alternative.**
+
+[![PyPI](https://img.shields.io/pypi/v/lmscan?color=blue)](https://pypi.org/project/lmscan/)
+[![License](https://img.shields.io/badge/license-Apache--2.0-green)](LICENSE)
+[![Python](https://img.shields.io/pypi/pyversions/lmscan)](https://pypi.org/project/lmscan/)
+[![Tests](https://img.shields.io/badge/tests-96%20passed-brightgreen)]()
+
+> GPTZero charges $15/month. Originality.ai charges per scan. Turnitin locks you into institutional contracts.
+>
+> **lmscan is free, open-source, works offline, and tells you _which_ model wrote the text.**
+
+```
+$ lmscan "In today's rapidly evolving digital landscape, it's important
+to note that artificial intelligence has become a pivotal force in
+transforming how we navigate the complexities of modern life..."
+
+🔍 lmscan v0.1.0 — AI Text Forensics
+══════════════════════════════════════════════════
+
+  Verdict:     🤖 Likely AI (77% confidence)
+  Words:       184
+  Sentences:   10
+  Scanned in 0.01s
+
+┌────────────────────────────┬──────────┬────────────────────┐
+│ Feature                    │ Value    │ Signal             │
+├────────────────────────────┼──────────┼────────────────────┤
+│ Burstiness                 │ 0.07     │ 🔴 Very low (AI)    │
+│ Sentence length variance   │ 0.27     │ 🟡 Below average    │
+│ Slop word density          │ 20.7%    │ 🔴 High (AI)        │
+│ Transition word ratio      │ 2.2%     │ 🟡 Elevated         │
+│ Readability consistency    │ 0.00     │ 🔴 Very low (AI)    │
+│ ...                        │          │                     │
+└────────────────────────────┴──────────┴────────────────────┘
+
+🔎 Model Attribution
+  1. GPT-4 / ChatGPT    62% — "delve", "tapestry", "beacon", "landscape" (×2), +19 more
+  2. Claude (Anthropic)  13% — "robust", "nuanced", "comprehensive"
+  3. Gemini (Google)      9% — "furthermore", "additionally"
+
+⚠️  Flags
+  • Very low burstiness (0.07) — AI text is more uniform in complexity
+  • High slop word density (20.7%) — contains known AI vocabulary markers
+```
+
+## Install
+
+```bash
+pip install lmscan
+```
+
+**Zero dependencies.** Works with Python 3.9+. No API keys. No internet. No GPU.
+
+## Usage
+
+```bash
+# Scan text directly
+lmscan "Your text here..."
+
+# Scan a file
+lmscan document.txt
+
+# Pipe from stdin
+cat essay.txt | lmscan -
+
+# JSON output (for scripts and CI)
+lmscan document.txt --format json
+
+# Per-sentence breakdown
+lmscan document.txt --sentences
+
+# CI gate: fail if AI probability > 50%
+lmscan submission.txt --threshold 0.5
+```
+
+### Python API
+
+```python
+from lmscan import scan
+
+result = scan("Text to analyze...")
+
+print(f"AI probability: {result.ai_probability:.0%}")
+print(f"Verdict: {result.verdict}")
+print(f"Confidence: {result.confidence}")
+
+# Which model wrote it?
+for model in result.model_attribution:
+    print(f"  {model.model}: {model.confidence:.0%}")
+    for evidence in model.evidence[:3]:
+        print(f"    → {evidence}")
+
+# Per-sentence analysis
+for sentence in result.sentence_scores:
+    if sentence.ai_probability > 0.7:
+        print(f"  🤖 {sentence.text[:60]}... ({sentence.ai_probability:.0%})")
+```
+
+### Scan entire directories
+
+```python
+from lmscan import scan_file
+import glob
+
+for path in glob.glob("submissions/*.txt"):
+    result = scan_file(path)
+    print(f"{path}: {result.verdict} ({result.ai_probability:.0%})")
+```
+
+## How It Works
+
+lmscan uses **12 statistical features** derived from computational linguistics research to distinguish AI-generated text from human writing:
+
+| Feature | What it measures | AI signal |
+|---------|-----------------|-----------|
+| **Burstiness** | Variance in sentence complexity | AI text is unusually uniform |
+| **Sentence length variance** | How much sentence lengths vary | AI produces uniform lengths |
+| **Vocabulary richness** | Type-token ratio (Yule's K corrected) | AI reuses words more |
+| **Hapax legomena ratio** | Fraction of words appearing once | AI has fewer unique words |
+| **Zipf deviation** | How word frequencies follow Zipf's law | AI deviates from natural distribution |
+| **Readability consistency** | Flesch-Kincaid variance across paragraphs | AI maintains constant readability |
+| **Bigram/trigram repetition** | Repeated word pairs and triples | AI repeats phrase structures |
+| **Transition word ratio** | "however", "moreover", "furthermore"... | AI overuses transitions |
+| **Slop word density** | Known AI vocabulary markers | "delve", "tapestry", "beacon"... |
+| **Punctuation entropy** | Diversity of punctuation usage | AI is more predictable |
+
+Each feature produces a signal via sigmoid transformation. The weighted combination produces the final AI probability.
+
+### Model Fingerprinting
+
+lmscan includes vocabulary fingerprints for 5 major LLM families:
+
+| Model | Distinctive markers |
+|-------|-------------------|
+| **GPT-4 / ChatGPT** | "delve", "tapestry", "landscape", "leverage", "multifaceted", "it's important to note" |
+| **Claude (Anthropic)** | "certainly", "I'd be happy to", "straightforward", "I should note" |
+| **Gemini (Google)** | "crucial", "here's a breakdown", "keep in mind" |
+| **Llama / Meta** | "awesome", "fantastic", "hope this helps" |
+| **Mistral / Mixtral** | "indeed", "moreover", "hence", "noteworthy" |
+
+Attribution uses weighted vocabulary matching, phrase detection, and hedging pattern analysis.
+
+## Accuracy & Limitations
+
+**What lmscan is good at:**
+- Detecting text with strong AI stylistic patterns
+- Identifying which model family generated text
+- Scanning at scale (thousands of documents) with zero cost
+- Providing explainable evidence (not a black box)
+
+**What lmscan cannot do:**
+- Detect AI text that has been manually edited or paraphrased
+- Work reliably on very short text (<50 words)
+- Detect AI text in non-English languages (English-only for now)
+- Replace human judgment — use as a signal, not a verdict
+
+**This is statistical analysis, not a neural classifier.** It detects stylistic patterns, not watermarks. It works best on unedited LLM output and degrades gracefully on edited text.
+
+## CI Integration
+
+### GitHub Actions
+
+```yaml
+- name: AI Content Check
+  run: |
+    pip install lmscan
+    lmscan submission.txt --threshold 0.7 --format json
+```
+
+### Pre-commit
+
+```yaml
+repos:
+  - repo: https://github.com/stef41/lmscan
+    rev: v0.1.0
+    hooks:
+      - id: lmscan
+        args: ["--threshold", "0.7"]
+```
+
+## Research Background
+
+lmscan's approach is informed by published research on AI text detection:
+
+- **DetectGPT** (Mitchell et al., 2023) — perturbation-based detection using log probability curvature
+- **GLTR** (Gehrmann et al., 2019) — statistical visualization of token predictions
+- **Binoculars** (Hans et al., 2024) — cross-model perplexity comparison
+- **Zipf's Law in NLP** — word frequency distributions differ between human and AI text
+- **Stylometry** — decades of authorship attribution research applied to AI forensics
+
+lmscan takes the statistical intuitions from these papers and implements them as lightweight, dependency-free heuristics that work without requiring a reference language model.
+
+## FAQ
+
+**Q: Is this as accurate as GPTZero?**
+A: GPTZero uses neural classifiers trained on labeled data. lmscan uses statistical heuristics. GPTZero is more accurate on edge cases; lmscan is free, offline, and explainable. Use both if accuracy matters.
+
+**Q: Can students use this to evade AI detection?**
+A: lmscan shows which features trigger detection, which could help someone understand why text reads as AI-generated. This is by design — understanding AI writing patterns makes everyone a better writer. The same information is available in published research papers.
+
+**Q: Does it work on non-English text?**
+A: Currently English-only. The slop word lists and transition word lists are English-specific. Statistical features (entropy, burstiness) work across languages but haven't been calibrated.
+
+**Q: Does it phone home?**
+A: No. Zero network requests. No telemetry. No API keys. Everything runs locally.
+
+**Q: How is model attribution possible without running the model?**
+A: Each LLM family has characteristic vocabulary biases. GPT-4 loves "delve" and "tapestry". Claude says "I'd be happy to". These are statistical fingerprints — not guaranteed attribution, but strong signals.
+
+## See Also
+
+- [reverse-SynthID](https://github.com/aloshdenny/reverse-SynthID) — Reverse-engineering Google's image watermarking
+- [vibesafe](https://github.com/stef41/vibesafe) — AI code safety scanner
+- [injectionguard](https://github.com/stef41/injectionguard) — Prompt injection detection
+- [vibescore](https://github.com/stef41/vibescore) — Grade your vibe-coded project
+
+## License
+
+Apache-2.0

lmscan-0.1.0/pyproject.toml

@@ -0,0 +1,71 @@
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[project]
+name = "lmscan"
+version = "0.1.0"
+description = "Detect AI-generated text and fingerprint which LLM wrote it. Open-source GPTZero alternative. Zero dependencies, works offline."
+readme = "README.md"
+license = {text = "Apache-2.0"}
+requires-python = ">=3.9"
+authors = [{ name = "Zacharie B" }]
+keywords = [
+    "ai-detection",
+    "ai-text-detection",
+    "gptzero",
+    "llm",
+    "chatgpt",
+    "claude",
+    "gemini",
+    "text-forensics",
+    "plagiarism",
+    "content-detection",
+]
+classifiers = [
+    "Development Status :: 4 - Beta",
+    "Environment :: Console",
+    "Intended Audience :: Developers",
+    "Intended Audience :: Education",
+    "Intended Audience :: Science/Research",
+    "License :: OSI Approved :: Apache Software License",
+    "Programming Language :: Python :: 3",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+    "Programming Language :: Python :: 3.13",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+    "Topic :: Text Processing :: Linguistic",
+    "Typing :: Typed",
+]
+dependencies = []
+
+[project.scripts]
+lmscan = "lmscan.cli:main"
+
+[project.urls]
+Homepage = "https://github.com/stef41/lmscan"
+Repository = "https://github.com/stef41/lmscan"
+Issues = "https://github.com/stef41/lmscan/issues"
+Changelog = "https://github.com/stef41/lmscan/blob/main/CHANGELOG.md"
+
+[tool.hatch.build.targets.wheel]
+packages = ["src/lmscan"]
+
+[tool.pytest.ini_options]
+testpaths = ["tests"]
+addopts = "-v --tb=short"
+
+[tool.ruff]
+target-version = "py39"
+line-length = 99
+
+[tool.ruff.lint]
+select = ["E", "F", "W", "I", "N", "UP", "B", "SIM", "TCH"]
+
+[tool.mypy]
+python_version = "3.9"
+strict = true
+warn_return_any = true
+warn_unused_configs = true

lmscan-0.1.0/src/lmscan/__init__.py

@@ -0,0 +1,16 @@
+from __future__ import annotations
+
+__version__ = "0.1.0"
+
+from .scanner import scan, scan_file
+from ._types import ScanResult, TextFeatures, SentenceScore, ModelMatch
+
+__all__ = [
+    "scan",
+    "scan_file",
+    "ScanResult",
+    "TextFeatures",
+    "SentenceScore",
+    "ModelMatch",
+    "__version__",
+]