Benchmarks

Kernel benchmarks exercise the engine on synthetic graphs. GTFS benchmarks run it on real public-transport data. Phases cover disk persistence, write pressure and vector search. Secret scope measures what sealing a branch costs. Each card says the kernel version its numbers were measured on.

.me kernel benchmarks

Kernel · 4.2.1Kernel benchmarksBenchmarks 1–11 and the fire test: scaling, fan-out, throughput, explain, secret scopes. Re-run on 4.2.0 (2026-10-10) next to the earlier builds on the same machine; cost timeline from 3.9.1 to 4.2.1 (on npm).

GTFS benchmarks

GTFS · 4.2.0GTFSReal transit feeds. Madrid (GTFS-Madrid-Bench), re-run on this.me 4.2.0.

Phase benchmarks

Phases · 3.6.4PhasesPhase 2: disk persistence and write pressure. Phase 3: exact vs IVF vector search (23.2× at 100k). Measured on 3.6.4 / 3.7.0, not re-run on 4.x.

Secret scope benchmarks

Secret scope · 4.2.0Secret scope & scalingSecret vs public branches: write and first-read cost across node counts, push vs pull, and the branch-scale run.