Replace list_repo_files() with per-protein lazy path probe

Fix event-loop blocking: make DB endpoints sync + raise timeouts

Add precomputed ML mechanistic labels as parquet (ml_preds/)

Add RaSP ΔΔG JOIN + proteome-wide download script

Fix esm1b fallback: force eager validation to catch httpfs 404

Add ESM1b/VESM++ JOIN to backend get_db() + fix hf_index scoping

fix: add huggingface_hub dep + httpfs fallback + cache bust

fix: add huggingface_hub dep + httpfs fallback + cache bust

refactor: serve parquet from local disk — baked in at build time

fix: persistent httpfs conn + skip INSTALL on warm requests

fix: hive_partitioning=true on direct-path reads — protein_id column was missing

fix: /api/info returns instantly on HF — no glob scan ever

fix: HF file index + direct-path reads — eliminates 429 rate-limit errors

fix: non-blocking startup cache — don't await count on 216M rows

Upload backend/__init__.py with huggingface_hub

Upload backend/requirements.txt with huggingface_hub

Upload backend/main.py with huggingface_hub