Privacy Policy

Last updated: September 2, 2026 — rewritten for 2.0; updated for 2.1's default-on sharing, for 2.2, which stopped collecting country and language, for 2.5's top-scores board, for 2.7.1, which cut that board to one row per hardware profile, for 2.8, which added each scene's raw stable load and a coarse display-pacing class to the sample, for 2.9, which added each scene's measurement flags and slowdown ratio and put thermally-flagged runs on the board, and for 2.11's leaderboard page, which shows up to 200 of those rows and older benchmark versions (see the changelog).

Short version: the screen tests run entirely in your browser and send nothing anywhere. The one thing this site can record is a single anonymous sample when you finish a scored benchmark run — and you're shown exactly what's in it, on screen, before the run starts, with a skip button right next to it.

What never leaves your device

The dead pixel test, gradient test, test patterns, device readout, and the effects gallery are local. No requests, no logging by us, no analytics scripts. Free-play and Classic benchmark runs don't submit anything either.

The one thing we collect, and when

When a scored Standard benchmark run completes, your browser sends one row to our stats service. Sharing is on by default: the start card that appears before every run shows a checked box and a live preview of the exact values your device would send — uncheck it and the run submits nothing, and that choice is remembered. This is the complete field list; nothing else is sent, ever:

• Benchmark result: total score, per-scene subscores with each scene's raw maximum stable load (the workload number already shown on your results sheet), whether that scene hit its time cap, its measurement flags ("drift" for a device that slowed as it warmed, "unconfirmed" or "floor" for a level that couldn't be proven) how much slower the end-of-scene check ran than the start — a single ratio like 1.3 — and whether that scene's number came from its one automatic re-measurement, plus benchmark version, profile, and whether the run completed cleanly, throttled, or degraded. All of this describes the run, not you; every field is already on your results sheet
• Screen numbers: resolution in pixels, pixel ratio, refresh rate bucketed to common tiers (60 / 75 / 90 / 120 / 144 / 165+), and a display-pacing class — "fixed", "adaptive", or "unknown" — describing how the display returns frames, so scores from different display types can be compared fairly
• Coarse hardware: OS family (e.g. "macOS"), browser family and major version (e.g. "Chrome 139"), GPU family normalized to a coarse label (e.g. "apple-m"), CPU cores and memory in buckets
• Device context: whether the device ran on battery or plugged in, its form factor (phone / tablet / desktop), and — on Android only — a coarse device family like "Galaxy S24" shared by millions of units; rare models are recorded as a brand bucket or "other", never as an exact model string, and other platforms record "none"
• Date: day precision only. We stopped recording country and site language entirely in version 2.2 — samples carry neither.

These rows power the public hardware benchmarks — medians over groups of at least 30 samples — and the top-scores board on the performance test page, which lists the best recent fully-measured runs individually (complete, and from 2.9 also thermally-throttled runs, which the board marks as such) — at most one row per coarse hardware profile, so one device running again and again can't fill the board: a score, and the same coarse families shown above. The leaderboard page shows the same rows to a depth of 200, for the last 30 days or all time, and for earlier benchmark versions as long as we still hold their samples. Nothing finer-grained than the field list is ever published, because nothing finer-grained exists.

What we will never collect

No IP addresses stored, no raw user-agent strings, no canvas or audio fingerprints, no stable device identifiers, no names, no emails, no precise timestamps, no free-text from your browser. There is no account system. Our own code sets no cookies; the only things it keeps in your browser's local storage are your own preferences (theme, your scores, your submit-or-skip choice, and which help cards you've dismissed), and they stay there. The one third-party service we use is described below.

Rate limiting

Our stats endpoint checks request rates by IP address to keep junk out of the dataset. The IP is used transiently for that check and is not written to the database or to logs we keep.

Retention

Raw samples are kept for 24 months, then deleted. Aggregated statistics (medians and percentiles per hardware family — never individual rows) are kept indefinitely and power the public stats pages.

No selling, no sharing

We don't sell, rent, or share the dataset with anyone. What we publish is aggregates with minimum sample-size thresholds, plus a top-scores board and a leaderboard listing the best runs, one per coarse hardware profile — each shown exactly as coarse as the field list above: a score, and hardware families shared by millions of devices. No published row contains anything to identify you with, because no stored row does.

Third parties

The site is served through Cloudflare, which processes requests the way any hosting provider does (their privacy policy applies to that).

We use Google Analytics 4 to understand aggregate site usage — which pages get visited, roughly where from, on what kinds of devices. It loads only after the page has finished rendering, and it may set cookies or similar identifiers governed by Google's privacy policy. We don't send it your benchmark results, and the anonymous stats dataset described above is completely separate from it. Blocking analytics with a content blocker breaks nothing on this site.

The optional gallery soundtracks stream from areal.me, another site we operate, served through the same Cloudflare setup as this one. Nothing is fetched from it until you switch the music on in the effects gallery, and turning it on sends an ordinary file request — no cookies, no identifiers, nothing about you beyond what any download carries.

We run no advertising scripts.

Changes

If this policy changes, the change is dated here and noted in the changelog. We won't quietly expand what's collected — a new field means a new consent card that says so.

Contact

Questions about any of this: admin at screentest.pro.