Blog
Announcements and notes from building the thing. RSS
A thousand runs of benchmark 2.5: what the hardware actually scores
1,065 runs in two weeks, charted: where GeForce, Radeon, Apple and Intel land, why Macs win two scenes outright, and what a battery really costs.
2026-09-17Same Mac, five browsers, five scores
One M2 Max, five browser engines, scores from 6,275 to 12,801 — what a browser benchmark actually measures, and what we changed because of it.
2026-09-11Twenty-four shaders, and the sliders to go with them
The effects gallery's Shader tier grew to 24 scenes, and every one now has sliders of its own. How we picked them, and why three scenes ask before they play.
2026-09-03Fifty-five checks, one click each
The homepage now shows every field, ramp and pattern as a chip that goes fullscreen in one click. Why we changed it, what's new, and what the numbers said.
2026-09-02Why Screen-Test a New Device Before You Buy
A Hong Kong Apple Store field test exposed a scoring flaw—and shows how to check touch, color, dead pixels, and performance before you buy.
2026-08-30The effects gallery got a soundtrack
Thirty original tracks, one per take on each of the 13 scenes. Open a scene, tap ♪, go fullscreen — it makes a genuinely good screensaver.
2026-08-23Building a browser benchmark that can't hang
Engineering notes from benchmark 2.0: why we search for load instead of counting FPS, and how a state machine keeps 150 seconds of GPU abuse escapable.
2026-08-23Screen Test Pro 2.0: same URLs, new everything
The rebuild is live: a two-tier benchmark, an instrument-style redesign, proper tool pages, and consent-first hardware stats.