VecForge 2.0 experiment results (JSON), from the experiment runs described in the paper.

results.json   E1 to E7: rank preservation of the reranker (E1), latency (E2), BM25 indexing (E3), namespace recall (E4), reopening (E5), search (E6) and memory (E7).
results2.json  E8 to E18:
  E8   SciFact retrieval quality: 5,183 documents, 809 training and 300 development queries, bootstrap confidence intervals and significance tests
  E9   encryption overhead
  E10  deletes
  E11  filters
  E12  ingestion across file formats (txt, md, html, docx, pdf)
  E13  memory deduplication
  E14  MCP interface
  E15  REST interface
  E16  security overhead
  E17  latency and time per stage
  E18  threads and concurrency

environment: CPU, cores, Python, NumPy and FAISS versions, and the baseline commit of the 1.x code. Timings were measured on a 4-core virtual machine, so compare ratios rather than absolute times.
