requirements.txt

# Runtime dependencies.
pydantic>=2          # schema / OrganoidProtocol contract
lxml>=5              # Tier-0 JATS parsing (pipeline/tier0_extract.py)
Pillow>=10           # Tier-2 figure handling (pipeline/tier2_vision.py)
datasette>=0.65      # serve layer (serve/run.sh, public deploy) — was missing (issue #5)
requests>=2.28      # Europe PMC discovery + SRI grounding HTTP (pipeline/discover_candidates.py, ground.py)
numpy>=1.24         # R3 semantic-retrieval vectors (pipeline/semantic_index.py; imported by its tests)

# R3 semantic-retrieval EXTRAS — needed only to `build`/`eval` the index, NOT for unit
# tests (semantic_index.py imports these lazily inside functions). For embedding/eval:
#   pip install sentence-transformers scikit-learn
# sentence-transformers>=2.2   # all-MiniLM-L6-v2 dense embeddings
# scikit-learn>=1.3            # TF-IDF lexical baseline in the eval