Open methodology

How we reach a verdict

Everything runs in your browser. No data leaves your device unless you press Share report. The logic below is complete — nothing is hidden.

Why is this different?

Speedtest measures one nearby server. This test checks nearby routes, far routes, and the apps you use — then shows what works and what doesn't.

Speedtest

My ISP Sucks

One test covers

One nearby server

One test covers

Nearby routes, far routes, your apps

Download, upload, ping

Yes

Download, upload, ping

Yes

Streaming, games, cloud

Estimated from the number

Streaming, games, cloud

Checked on those services

What works and what doesn't

Not shown

What works and what doesn't

Shown

Who is responsible

Not named

Who is responsible

Wi‑Fi, ISP, or one service

You can send

The number

You can send

A complaint

Not affiliated with Ookla or Speedtest.

Run this test

1 · What we probe

Four concentric layers, from your device outward:

Local gateway (preflight)

10 sequential HEAD requests to the nearest Cloudflare PoP (600 ms timeout each). Measures loss and jitter before anything else. If ≥ 2 time out or std-dev > 30 ms → path marked unstable.

Same-country nodes (near)

AWS S3 endpoints inside your country (matched by IP geolocation country code). Low score here = domestic ISP or routing issue.

International nodes (far)

16 AWS S3 regional endpoints worldwide (N. Virginia, Frankfurt, Tokyo, Singapore…). High score here while near nodes are fine = international gateway problem.

Service endpoints (profile)

15–25 HEAD requests to the services in your chosen profile (streaming / gaming / tech). Detects peering and blocking issues invisible to raw ping.

2 · Node health score

Each node gets a score from 0 (perfect) to 1 (broken). Near and far averages are kept separate.

score = 0.7 × (packetLoss / 100)
      + 0.3 × clamp((actualMs − expectedMs) / (expectedMs × 3), 0, 1)

expectedMs = max(10, distanceKm / 50)   ← speed-of-light RTT baseline

Example — Tel Aviv user → N. Virginia node (9 500 km)

expectedMs = max(10, 9 500 ÷ 50) = 190 ms

actualMs = 219 ms, packetLoss = 0 %

score = 0 + 0.3 × clamp((219 − 190) / (190 × 3), 0, 1) = 0.015 → healthy

nearScore < 0.25

Near: healthy

farScore > 0.35

Far: triggers Rule 6

farScore > 0.45

Far: degraded (Rule 5b)

3 · Service outlier

outlier if:  status = "fail"
         OR  status = "slow"  AND  latencyMs ≥ cohortMedian × 2.5

slow = probe latency above 1 000 ms
ok under that line is not an outlier

Example

Netflix 95 ms ✓ YouTube 88 ms ✓ Twitch 1 240 ms

cohortMedian = 88 ms → 1 240 ≥ 88 × 2.5 (220) → Twitch is an outlier

4 · Verdict rule chain

First match wins. Open DevTools → Console on any result page to see every rule that fired or was skipped.

Offline

connectivity = fail, and no other check reached the internet

→ down / local_issue

Local path unstable

Wi-Fi or cellular AND (preflight unstable OR jitter > 40 ms) AND farScore < 0.40

→ local_issue

DNS broken

device is online AND dns = fail

→ isp_issue

Specific service outlier

core healthy AND 1 ≤ outliers ≤ 40 % of services

→ service_issue

Most services degraded

core healthy AND badServiceFraction ≥ 50 %

→ isp_issue (peering)

Intl. traffic blocked

preflight loss = 0 AND nearScore < 0.25 AND farScore > 0.45 AND ≥ 3 far nodes

→ isp_issue (gateway)

Core path / far degraded

any test = fail OR farScore > 0.35

→ isp_issue

Soft degradation

any test = warn (no harder rule fired)

→ local_issue or isp_issue (medium)

✓

All clear

→ healthy

5 · Plan delivery

ratio = measuredDown / promisedDown

≥ 0.80  →  "As advertised"    ✓
≥ 0.50  →  "Underdelivering"  ~
 < 0.50  →  "Far below plan"   ✗

Only annotates the report — does not change the culprit verdict.

6 · Privacy

All probes run in your browser. Nothing is sent to our server unless you press Share report.

Shared records store: verdict, scores, ISP name, country, medium. No IP addresses are stored.