mika's open learning notebook

Performance starts with asking what must exist

A performance pass is not just shaving milliseconds from code. Often the better question is whether a piece of work needs to happen at all.

In a 3D library, thousands of unique meshes, materials, image textures, labels, and updates can become expensive quickly. A convincing world can instead reuse geometry, reduce distant detail, defer expensive content, and keep only nearby interactions fully alive.

Measure the workflow, not one function

Performance is experienced across the main loop: loading, movement, targeting, opening content, closing it, and continuing to explore. An optimization that makes startup faster but introduces stutter during navigation is not necessarily a win.

Cleanup matters

Interactive features create resources: event listeners, timers, audio, geometry, materials, textures, windows, and state subscriptions. If ownership is unclear, resources survive longer than intended and bugs accumulate.

This connects directly to what I learned building Mochi: long-running systems need explicit ownership and cleanup paths.

What I can do now

I can look for architectural causes of poor performance, distinguish visual density from actual workload, and evaluate optimization against the user’s complete interaction loop.