PDF features with 30%% dropout: page_count, text_density, fonts, aspect_ratio from PosterSentry

1057-dim model: +8 PDF features (page_count, text_density, fonts, aspect_ratio), content-skip heuristic

Update: +1425 bioRxiv preprint samples, 3x clean window, bioRxiv boilerplate strip

Update head_toxicity.npz: 3x clean window, bioRxiv boilerplate stripping, 10K abstract search

Update head_ai_detect.npz: 3x clean window, bioRxiv boilerplate stripping, 10K abstract search

Update head_doc_type.npz: 3x clean window, bioRxiv boilerplate stripping, 10K abstract search

Update embedding model cache

Update head_toxicity.npz: merged 5-class doc_type (non_research), dual embeddings, enriched training data

Update head_ai_detect.npz: merged 5-class doc_type (non_research), dual embeddings, enriched training data

Update head_doc_type.npz: merged 5-class doc_type (non_research), dual embeddings, enriched training data

Update README.md

Update README: 5-class doc_type with literature_review, updated metrics

Update head_doc_type.npz (retrained with full-text review data)

Update toxicity: 83.9% accuracy

Update ai_detect: 84.2% accuracy (RAID + ChatGPT-Abstracts)

Update doc_type: 5-class balanced at 15K/class, 96.0% accuracy, 5/5 smoke tests

Update toxicity head (84.2% accuracy)

Update ai_detect head (84.5% accuracy)

Update doc_type head: 5-class with literature_review (96.1% accuracy)

Add config.json for HuggingFace download tracking

Fix gate: only scientific_paper passes

Fix gate: only scientific_paper passes

Fix gate logic docs: only scientific_paper passes, posters are blocked

Add pip install from GitHub URL

Updated README — logo, architecture diagram, real training data, gate logic

Add PubGuard logo

Updated with real poster data support + poster pass-through gate

Updated with real poster data support + poster pass-through gate

Updated with real poster data support + poster pass-through gate

Updated with real poster data support + poster pass-through gate

Retrained head_toxicity.npz — real poster data, no synthetic junk

Retrained head_ai_detect.npz — real poster data, no synthetic junk

Retrained head_doc_type.npz — real poster data, no synthetic junk

Clean gate script — no errors module dependency

Remove error module imports from __init__.py

Remove error code references from README

Remove errors.py (internal only)

Remove ERRORS.md (internal only)

Upload ERRORS.md with huggingface_hub

Upload pyproject.toml with huggingface_hub

Upload scripts/pubguard_gate.py with huggingface_hub

Upload scripts/train_pubguard.py with huggingface_hub

Upload src/pubguard/classifier.py with huggingface_hub

Upload src/pubguard/train.py with huggingface_hub

Upload src/pubguard/cli.py with huggingface_hub

Upload src/pubguard/config.py with huggingface_hub

Upload src/pubguard/text.py with huggingface_hub

Upload src/pubguard/__init__.py with huggingface_hub

Upload src/pubguard/data.py with huggingface_hub

Upload src/pubguard/errors.py with huggingface_hub