Voxiesz/fluctlightdb-benchmarks

FluctlightDB — Frozen Benchmark Results

Official frozen metrics for the FluctlightDB research paper (June 2025).

Files

File	Description
results.json	Full benchmark output — LoCoMo, BEIR SciFact, FAMB

Key numbers

• LoCoMo evidence recall: 98.1% (1925/1982 gold spans, k=150, hybrid)
• BEIR SciFact nDCG@10: 0.645 (index mode, ties Chroma)
• FAMB macro: 98% index / 97% agent

Reproduce

git clone https://github.com/voxmastery/FluctlightDB.git
cd FluctlightDB
# See benchmarks/README.md and docs/BENCHMARKS.md

Paper

• LaTeX: https://github.com/voxmastery/FluctlightDB/tree/main/papers/arxiv-v1
• Card: https://huggingface.co/Voxiesz/fluctlightdb-paper

Citation

Use CITATION.cff from the main repository.