What to Load-Test Before You Need To

Scale breaks assumptions: O(n²) loops, missing indexes, unbounded queues, synchronous chains, hot keys in cache.

Load-test the paths that matter economically — checkout, auth, search — not only the homepage.

Watch p95 and error rate, not average latency. Averages lie when tails kill users.

Scaling problems are often validation: people want the product. Your job is to make wanting it survivable.

Design for the storm while seas are calm — caches, backpressure, idempotency, graceful degradation.

— JV · Dark Heart Labs.