Evidence

Reproducible benchmarks

Interpret and reproduce Sightglass ingestion, query, and storage measurements.

The benchmark suite builds the real server, creates isolated temporary databases, and records raw samples before generating summary tables and SVG charts. It measures HTTP ingestion, representative dashboard queries, and SQLite storage footprint.

The current reference was measured 12 September 2026 on Node.js 24.18.0, Windows x64, an Intel Core i5-10300H with 8 logical CPUs, and 16 GB RAM.

PathMedianp95 sample
In-memory SQLite ingest2,713 occurrences/s2,833 occurrences/s
SQLite WAL ingest2,301 occurrences/s2,437 occurrences/s
Exact meter ledger8,867 meters/s10,357 meters/s
HTTP + JSON + validation + WAL1,586 occurrences/s1,758 occurrences/s

Measured ingestion throughput

Dashboard query over 25,000 operationsMedianp95
Overview summary5.81 ms11.52 ms
Occurrence list0.36 ms0.50 ms
Occurrence detail0.06 ms0.23 ms
Database ranking0.22 ms0.27 ms
Dependency ranking0.24 ms0.32 ms
Usage totals0.97 ms1.54 ms

Measured dashboard query latency

Run it from the main source repository:

npm run benchmark

The checked-in evidence is a regression reference, not a universal capacity promise. Results depend on CPU, filesystem, container limits, payload shape, cardinality, retention, and concurrent application traffic. Load-test with production storage and resource limits before sizing a deployment.

The machine-readable result includes every raw sample, descriptive statistics, settings, and environment details.