Evaluator
What is real here?
Focus on supported surfaces, capability boundaries, and benchmark honesty.
Treat the site like a compact academy: different readers need different routes through the same material.
| Track | Best for | Read in this order |
|---|---|---|
| Evaluator | Interviewers, maintainers, skeptical reviewers | Homepage → Guided Start → Architecture Overview → Benchmarks |
| API reader | Engineers deciding whether to integrate | Quick Start → API Reference → Browser Support |
| DSP learner | Readers refreshing FFT theory | Tutorials → Architecture Overview → Academic Papers |
| Contributor | People likely to modify code or docs | Architecture Overview → Architecture Decisions → API Contract → Contributing |
Evaluator
Focus on supported surfaces, capability boundaries, and benchmark honesty.
API reader
Focus on data contracts, installation cost, browser support, and failure paths.
DSP learner
Focus on radix-2, Hermitian compression, GPU mapping, and external references.
Contributor
Focus on API/documentation truth, architecture decisions, and the low-maintenance repo posture.
Do not read “GPU-accelerated” as a blanket statement for the entire library. The core FFT path is GPU-capable, while spectrum analysis and image filtering remain CPU-only. The strongest reading path is the one that keeps that distinction in view from the beginning.