Scan a package
homebench

HydraDB-shaped bench.

Same CSV columns as examples/query_bench.rs in the HydraDB repo, so the same tooling reads Meridian numbers alongside the upstream numbers. The canonical file is also written to bench/out/cypher_bench.csv by pnpm bench.

CSVdisplay columns · query_shape · cold_open_query_p50_us · cold_open_query_p95_us · hot_p50_us · hot_qps · concurrent_qps · rows▸ download csv
query_shapecold_open_query_p50_uscold_open_query_p95_ushot_p50_ushot_qpsconcurrent_qpsrows

planner shape

Each row in the CSV is optimizer_plan-annotated. The shape column is what the planner reports for the headline traversal: a 6-hop reverse traversal over dependency edges with a service-type filter pushed below the index probe.

hot QPS · 1,800 / node

On the canonical 5K-node / 18K-edge fixture, a single Meridian worker sustains ~1,800 hot QPS against the local HydraDB instance, with p99 hot-query latency < 60 µs. Concurrent QPS scales to ~1,620 with 16 parallel clients.