research(nightly): symphonyqg — co-designed 1-bit graph quantization (SIGMOD 2025)#428

Draft

ruvnet wants to merge 13 commits intomainfrom

research/nightly/2026-05-07-symphonyqg

Commits on May 7, 2026

feat(nightly): add ruvector-symphonyqg — SymphonyQG co-designed graph+quantization (SIGMOD 2025)

claude
authored and
ruvnet
committed
fix(symphonyqg): padding-edges correctness + ADR rename + clippy/fmt

ruvnet
and
ruvnet
committed
style: cargo fmt drift cleanup in ruvector-graph + ruvector-hailo

ruvnet
and
ruvnet
committed
perf(symphonyqg): repack adjacency+codes into single per-vertex block (SOTA layout)

ruvnet
and
ruvnet
committed
test(symphonyqg): add 5 reviewer-flagged edge-case tests

ruvnet
and
ruvnet
committed
docs(ADR-193): honest measurements + structurally-delivered cache locality

ruvnet
and
ruvnet
committed
perf(symphonyqg): u64-chunked XNOR-popcount + Config::validate() wired up

ruvnet
and
ruvnet
committed
style(symphonyqg): cargo fmt — single-line from_raw_parts_mut call

ruvnet
and
ruvnet
committed
feat(symphonyqg): Vamana α-pruning refinement (DiskANN, NeurIPS 2019)

ruvnet
and
ruvnet
committed
feat(symphonyqg): wire Vamana into Config + measure +26.8pp recall at n=50K

ruvnet
and
ruvnet
committed
fix(symphonyqg-vamana): back-edge propagation + medoid entry point + honest docs

ruvnet
and
ruvnet
committed
feat(symphonyqg): rayon-parallel search_batch + 13.83× measured speedup

ruvnet
and
ruvnet
committed
docs(ADR-193): note iter-8 rayon search_batch (13.83× measured speedup)

ruvnet
and
ruvnet
committed