Five free instruments

A PageSpeed score is not a decision.

Value a millisecond, test a proposed fix, set a defensible budget, detect score theater and prove a before/after change — on your data, built on the same evidence standards as the audit and our published research. No made-up multipliers, no folklore numbers. No paywall, and no email gate on the calculators — Fix Simulator is the one that needs a free account, because it stores your runs to compare them.

Price my page speed Choose another tool

Our own pages, tested with the tool we point you at Why lab and field disagree

Start with the decision

What do you need to know next?

Five tools, five distinct decisions. Pick the question blocking the next action — not the metric with the loudest red score.

01Value speed 02Test a fix 03Set a budget 04Check integrity 05Prove a change

Decide what to ship

A fix and a budget answer different questions

First test whether the proposed change creates a measurable effect. Then turn the target into a constraint the next release cannot quietly break.

predicted vs. measuredExample output
Baseline median 4.20 s
After fix 3.56 s
PSI predicted −880 ms
Measured −640 ms

Repeated trials per variant separate an effect from ordinary Lighthouse jitter.

02 · Test a proposed fix Free · account

Fix Simulator

Stop guessing whether a change will help. Toggle a hypothetical fix and measure the real before-and-after with Google's own PageSpeed data — predicted vs. measured, side by side.

Input
URL + the change you're considering
Output
Predicted against measured effect
Access
Free, with a free account required — it stores your runs to compare them

Run a fix experiment

Evidence: Do Lighthouse's promised savings materialize? · controlled calibration pilot

critical-path budgetExample output
Total critical-path budget 247 KB
Profile Slow 4G · 150 ms RTT
HTML 19 KB
CSS 29 KB
LCP image 111 KB

At most 11 sequential request hops fit this target — one more leaves no time to move bytes.

03 · Set the delivery constraint Free

Performance budget calculator

A TCP-aware budget: how many kilobytes and requests you can afford before your target device blows past your speed goal — not a round number pulled from thin air. Optionally start from your URL: your real users' p75 for context, your measured TTFB prefilled.

Input
LCP target + connection profile
Output
Byte budget + affordable request hops
Artifact
Exportable Lighthouse budget.json

Build my performance budget

Evidence: The cost of a kilobyte · 210-run controlled experiment

Verify the evidence

A result is only useful if the page and the claim are real

Check whether Google received the same experience your visitors get — then work out how much repetition a before/after claim actually needs.

normal visitor vs. LighthouseExample output
Verdict No material mismatch
Probes 3 agree
Normal mobile 128 KB HTML
Lighthouse UA 126 KB HTML
Script tags 31 vs 30
Third-party hosts 7 vs 7

A discrepancy is a reason to inspect. The live tool shows the signals and their limits before it escalates to a verdict.

04 · Check measurement integrity Free

PageSpeed cloaking test

Is a site serving a faster page to Google's crawler than to real users? This probes for PageSpeed cloaking and shows the gap — yours, or a competitor's.

Input
Any public page URL
Output
Visitor-vs-Google evidence, with its limits
Standard
No accusation from one weak signal

Compare Google's view

Evidence: How many sites cheat PageSpeed? · top-2,000-site pilot

variance and powerExample output
Runs required per condition 25
At 80% power · α .05
Protocol · 25 runs detects δ = 5
Median-of-5 · 5 runs not enough

Bars are runs against the 25 this page needs at σ̂ = 5.0 points, judged by median as the verdict is. A median-of-five can stabilise an estimate. It cannot, by itself, prove a small change is real.

05 · Prove a before/after claim Free

Lighthouse required-runs calculator

“The score went up two points” usually isn't evidence. This measures your page's real run-to-run variance (a short PSI pilot, per metric — score, LCP, TBT, FCP, CLS), tells you how many runs a claim actually needs, writes the experiment protocol — and takes your before/after Lighthouse JSONs for a per-metric statistical verdict: real improvement, regression, or within noise.

Input
URL, a preset, or your measured σ
Output
Required n + the experiment protocol
Verdict
Improvement, regression, or within noise

Prove my before/after change

Evidence: How many Lighthouse runs prove a change? · analytic power model

Method, not magic

Every instrument exposes its evidence base

The previews make the output tangible. The linked research makes the assumptions, the study design, the raw data and the limitations inspectable. These are five independent evidence bases, not five stages of one process.

11 records Speed-to-revenue evidence, graded by research design 8 cases Predicted against observed, in a controlled pilot 210 runs Controlled byte and critical-path experiments 1,162 sites Measured in the cloaking-prevalence pilot Power model Per-metric sample size and uncertainty

When one answer is not enough

Five answers are useful. One prioritised plan is more useful.

The tools each answer one question. The expert audit answers all of them for the page that earns your money — every finding backed by your own markup, every fix a concrete code change. Curious how we know what we know? Read the research.

Get your audit How it works
What the free instruments do, and what the audit adds on top.
Free instrumentsExpert audit
Answer one decision at a timeAdjudicates every finding on the page
Self-service inputs and scenariosReads your markup, headers, scripts and critical path
One output or statistical verdictPrioritised code changes with a predicted range
You combine the conclusionsOne constraint-first implementation sequence
Free to run whenever you need themVerification re-runs included