PageSpeed Insights and Lighthouse report analyzer

What PageSpeed Insights puts in one response

PageSpeed Insights can return two fundamentally different kinds of evidence. Its lighthouseResult is a lab run: one controlled simulation with a recorded Lighthouse version, device environment, URL and fetch time. When enough Chrome User Experience Report data exists, loadingExperience and originLoadingExperience add aggregated real-user measurements. Google documents those fields separately in the PageSpeed Insights API v5 response.

The analyzer preserves that boundary. Lab category scores appear in one card. URL- and origin-level LCP, INP and CLS appear in another only when the JSON contains them. No formula combines the two. This matters because a single clean run can coexist with slow real-user devices, while a noisy lab run can coexist with acceptable field experience. They are clues from different populations, not two halves of one score.

Why a Lighthouse report needs its version and audit IDs

A Lighthouse JSON file is more useful than a screenshot because the structured report keeps every audit under a stable ID, records whether it passed, failed, was manual or did not apply, and can include numeric savings. Categories then reference weighted audits. This tool reads that relationship rather than trying to reconstruct a score from colorful gauges. Google’s own guide to Lighthouse results also distinguishes audit results from category groupings and weights.

Audit definitions change as Lighthouse evolves. The rule map here is deliberately bounded to major versions 10 through 13. An older or future report is not silently treated as current. Its category summary remains visible, but every version-bound rule is marked skipped. That is less exciting than guessing and much safer than recommending work against an audit that was renamed, removed or rescored.

Performance scores are queues, not universal grades

Lighthouse labels performance scores from 0–49 as poor, 50–89 as needing improvement and 90–100 as good, but Google also warns that scores vary because network routing, client hardware, browser extensions and background work vary. The uploaded score is therefore shown exactly as Lighthouse supplied it. The analyzer does not promise that turning a 62 into a 90 will produce a particular traffic or revenue change.

The action queue instead groups related failures: LCP discovery and rendering, main-thread blocking, layout stability, JavaScript and third parties, CSS, images, server delivery, accessibility and SEO fundamentals. Evidence retains each audit ID and its display value or reported savings. Start with the highest defensible queue item, make one coherent change and repeat an equivalent profile several times. A one-run score difference is not automatically proof that the change caused it.

Core Web Vitals use field thresholds at the 75th percentile

The published “good” field thresholds are LCP at 2.5 seconds or less, INP at 200 milliseconds or less and CLS at 0.1 or less, evaluated at the 75th percentile of page loads. The uploaded PageSpeed response already carries its field categories and percentiles, so the tool displays those values rather than manufacturing a new grade. See Google’s explanation of how Core Web Vitals thresholds are defined for the population and percentile logic.

Lighthouse cannot observe representative user interaction and therefore cannot measure real-user INP. Total Blocking Time is useful in the lab because long main-thread tasks often reveal code that will also make interaction harder, but it remains a diagnostic proxy. This report never converts TBT into predicted INP. If the JSON has no field INP, the correct output is “not available,” not a derived number.

Optional HTML evidence has a narrow job

Pasting View Source HTML can confirm a few source-visible facts. The analyzer counts image tags missing either dimension, checks for a viewport meta tag and counts external script hosts written directly into the markup. Those checks can make an audit easier to route to the right template or vendor owner. They cannot see a tag manager’s runtime injections, response headers, a service worker, external CSS rules or the browser’s actual request priority.

That limitation is written into the generated finding. A source-visible signal should be connected to the matching Lighthouse evidence before it becomes work. For example, an image without dimensions supports a layout-shift investigation only when the rendered element can actually move content. The source alone cannot establish that causal chain.

A deterministic report with an explicit rule ledger

Each rule says whether it ran, which supported audit IDs were present and how many findings it produced. Missing SEO or accessibility categories are skipped rather than treated as perfect scores. Manual accessibility audits are not counted as automated passes. A report with no findings therefore means only that none of the supported conditions crossed its threshold; it is not an accessibility, SEO or performance certificate.

Priority uses impact × confidence × reach ÷ effort, with every factor shown on each card. Lighthouse score and reported savings drive the evidence side. Your stack and implementation access can change the likely owner or effort, while a required third-party script changes the first action from removal to loading strategy. Download CSV for a backlog or Markdown for a review note. If the problem is search visibility rather than page delivery, move to the Search Console report analyzer or begin with the interactive Google ranking diagnosis.

Your Lighthouse JSON remains on this device

The browser reads the file into this tab and sends its text to a dedicated browser worker. Parsing and scoring happen locally; no report body is sent to CreatorValet, Google or an AI provider. The worker keeps large JSON work away from the page controls, and closing the tab discards both inputs and result. If you only need to inspect or repair an unfamiliar JSON structure first, use the JSON formatter and return with the original structured Lighthouse object intact.