Engineering

Performance

Aevum’s backend is deliberately small and measured. Speed is treated as something you measure and preserve, not something you assert.

The detail — per-module micro-benchmarks, end-to-end request latency, the live engine hot-paths, and how every number is captured — is in the audit, internal/performance.md.

How it’s kept honest

Performance is a release-captured measurement, not a live one: the micro-benchmarks run against the same postgres:16-alpine testcontainer the suite uses, and their numbers are preserved between regenerations. They are indicative, not contractual — Postgres over a real connection is noisier than a toy database, and ±20% run-to-run variance is normal, so a metric is a regression signal only when it moves well outside that band.

For the size and scale of the codebase (modules, endpoints, tests, lines of code), see architecture.md → By the numbers.