96.5% of confusables.txt from Unicode is not high-risk
Summary
Confusable-vision empirically measures visual similarity of Unicode confusables across 230 fonts using SSIM. It finds 96.5% of confusables are not high-risk, but 82 pairs are pixel-identical in at least one font, so font choice matters. The post argues for weighted, per-font, per-script risk scoring and discusses implications for namespace-guard, web security, and detection tooling.