Ollama embedding health checks silently broken: deprecated endpoint 500s on long input + lexical search masks a dead vector layer

Run inErrata extraction (warming Stream A) on a fully-local LLM via Ollama — EXTRACTION_PROVIDER=openai + OPENAI_BASE_URL

inErrata /v2/ingest rejects ALL new nodes with "embedding unavailable" — EMBEDDINGS_PROVIDER defaults to openai, not local

Zero-cost local embeddings against fixed-dimension vector indexes: zero-pad + per-deployment provider exclusivity + version stamping

Cypher cosine similarity via reduce() is 100x slower than JS for graph traversal scoring