Tested 2026-07-27 23:54:20 using Chrome 150.0.7871.186 (runtime settings)
This page is fast to paint, snappy to interact with, visually stable and follows best practices, but weak on user privacy.
Visual metrics
Google Web Vitals
The page has 1 image (out of 1) without a decoding hint. Add decoding="async" to non-critical images so the browser can decode them off the main thread.
The page ships 1 image (out of 1) in JPEG/PNG/GIF without a modern alternative. Wrap them in a <picture> with a <source type="image/avif"> or "image/webp" before the legacy <img>, or serve modern formats from your image pipeline directly. AVIF and WebP usually deliver 25–50% smaller files at the same quality.
The page has 3 CPU long tasks with the total of 668 ms. The total blocking time is 59 ms and 2 long tasks before first contentful paint with total time of 559 ms. However the CPU Long Task is depending on the computer/phones actual CPU speed, so you should measure this on the same type of the device that your user is using. Use Geckoprofiler for Firefox or Chromes tracelog to debug your long tasks.
Timings summary
| Metric | Min | Median | Mean | Max | Runs agree? |
|---|---|---|---|---|---|
| First visual change | 900 ms | 1.100 s | 1.044 s | 1.133 s | runs disagree ±11% |
| Last visual change | 1.633 s | 1.633 s | 1.633 s | 1.633 s | identical |
| Speed index | 1.261 s | 1.335 s | 1.320 s | 1.365 s | stable ±4% |
| Largest image | 900 ms | 1.100 s | 1.044 s | 1.133 s | runs disagree ±11% |
| Heading | 1.500 s | 1.533 s | 1.522 s | 1.533 s | stable ±1% |
| Largest contentful paint | 1.267 s | 1.300 s | 1.289 s | 1.300 s | stable ±1% |
| Last meaningful paint | 1.500 s | 1.533 s | 1.522 s | 1.533 s | stable ±1% |
| Visual readiness | 500 ms | 533 ms | 589 ms | 733 ms | runs disagree ±22% |
| Visual complete 85 | 1.500 s | 1.533 s | 1.522 s | 1.533 s | stable ±1% |
| Visual complete 95 | 1.500 s | 1.533 s | 1.522 s | 1.533 s | stable ±1% |
| Visual complete 99 | 1.633 s | 1.633 s | 1.633 s | 1.633 s | identical |
Run 2 SpeedIndex median
The red band on the timeline marks when the main thread was busy with long tasks: moments when the page could not paint or respond, however the video looks.
Download videoUse --filmstrip.showAll to show all filmstrips.
Click a frame to seek the video to that moment. The dot marks frames where the page changed visually.
The whole recording in one image: every pixel is tinted by how often it changed. Amber changed once, deep red changed ten or more times. The scale is fixed, so heatmaps from different runs compare directly.
Flat amber is a page that painted once and stayed put. Red speckle in the shape of the text means the same pixels repainted over and over, usually invisible rendering noise that inflates Last Visual Change and Speed Index. A solid red block is a genuinely restless area, like an ad or a carousel.
How smoothly Chrome delivered frames to the screen during the test. Dropped counts only frames Chrome flags as affecting smoothness. On mostly-static pages the longest gap includes idle time where there was simply nothing to draw.
Rendering was not meaningfully delayed. First paint landed at 1.084 s on a 424 ms first byte; none of the 1 render-blocking request held it up: it loaded after the paint. The largest paint followed at 1.284 s.
Style the browser had to recalculate before it could paint: many elements or a lot of time spent here points at expensive CSS on the critical rendering path. The recalculation count and the largest single one tell the two problems apart: one massive style change touching most of the page (few recalculations, the largest close to the total) needs different medicine than thousands of tiny ones (many recalculations, each tiny).
One large recalculation dominates. Restyling 908 elements cost 106 ms before the page painted (9.8 % of the time to First Contentful Paint), and a single recalculation of 833 elements accounts for 85 ms of it. Look for one class or inline-style change on a large container near the top of the DOM that restyles the whole subtree at once. The changes driving it are listed under Style invalidations below.
The recalculation work before first paint is 9.8 % of the time to First Contentful Paint.
Blocking requests delay both paints: the browser renders nothing until they arrive, so they push First Contentful Paint and everything after it, including the largest paint (LCP). Requests that block the parser inside the body stall everything after them in the document. Load JavaScript with defer or async, and keep non-critical CSS off the critical path with media attributes. Bars show when each request loaded on a clock that runs to the last paint (the dashed line is FCP, the clock ends at LCP); the gap after the bars is style and layout work.
| Request | When it loaded | Transfer | Download |
|---|---|---|---|
document | 509 ms | 31.2 KB | 508 ms |
index-34f340e24a.jsblocking | loaded after LCP, at 1.290 s | 9.2 KB | 108 ms |
gt-ie9-507b16b6be.jsblocks parser in body | loaded after LCP, at 1.293 s | 1.4 KB | 105 ms |
LCP resource | 163 ms | 18.8 KB | 163 ms |
0321 ms642 ms963 msFCP 1.084 sLCP 1.284 s |
How long Chrome spent testing each CSS selector against the page while recalculating styles. A selector with many match attempts and zero matches is evaluated against the document over and over for rules that never apply.
Not worth optimizing on this page. All selector matching cost 26 ms under the 7× CPU throttle, roughly 4 ms real time, and the worst single selector 2.33 ms. The tables are kept for reference.
| Selector | Time | Attempts | Matches |
|---|---|---|---|
.styled-select.js-enabled select option | 2.33 ms | 154 | 77 |
.hlist li:first-child::before | 1.71 ms | 694 | 13 |
a:active | 1.68 ms | 772 | 0 |
.banner * | 1.63 ms | 1 577 | 0 |
.langlist a | 1.49 ms | 386 | 346 |
.langlist > ul > li a | 1.08 ms | 386 | 344 |
.langlist > ul > li | 0.85 ms | 694 | 343 |
a | 0.81 ms | 772 | 772 |
.jsl10n-visible .jsl10n | 0.79 ms | 90 | 42 |
.hlist li::before | 0.14 ms | 694 | 692 |
| Selector | Attempts | Time |
|---|---|---|
.banner * | 1 577 | 1.63 ms |
.central-featured-lang :hover | 1 577 | 0.08 ms |
a:active | 772 | 1.68 ms |
a:hover | 772 | 0.12 ms |
.no-js .langlist > ul > li | 694 | 0.03 ms |
.banner-bottom .portal-banner-content-message p a:hover | 386 | 0.03 ms |
.banner-bottom .portal-banner-content-message p a | 386 | 0.01 ms |
abbr[title] | 135 | 0.01 ms |
.central-featured-lang .link-box:hover strong | 24 | 0.01 ms |
.banner-bottom .portal-banner-donate-button p strong | 24 | 0.00 ms |
Why the browser had to recalculate style and layout: every DOM/style change invalidates some set of nodes, and the same class or attribute churning over and over is the pattern to hunt. Counts, not milliseconds: the time is in the style recalculation numbers above.
Chase what keeps changing after the page is up. 339 style recalculations and 1 262 layout invalidations fired after first paint, on a page the user was already looking at. The biggest driver is pseudo invalid (6×). Start with the script under "Scripts causing invalidations" that keeps triggering it.
pseudo: invalid6×pseudo: checked2×pseudo: focus2×class: jsl10n-visible1×class: js-enabled1×class: no-js1×Invalidations while the largest paint was still pending, so these delayed it.
| Reason | Style recalculations | Layout |
|---|---|---|
| A node was inserted into the page | 156 | – |
| An element entered layout | – | 635 |
| An SVG element changed | – | 448 |
| Unknown | – | 2 |
| An element's size changed | – | 2 |
Churn on the fully rendered page: wasted work and possible jank, but no paint delay.
| Reason | Style recalculations | Layout |
|---|---|---|
| A node was inserted into the page | 35 | – |
| Related style rule | 125 | – |
| A pseudo-class state changed (:hover, :focus, …) | 10 | – |
| Attribute | 10 | – |
| Control | 1 | – |
| A stylesheet rule changed | 1 | – |
| JavaScript changed an inline styleSet a class instead of writing element.style, so the browser can batch the work. | 1 | – |
| An element entered layout | – | 126 |
| A style change forced layout | – | 46 |
| An element left layout | – | 3 |
The page being constructed and styled for the first render. Expected work, shown for scale.
| Reason | Style recalculations | Layout |
|---|---|---|
| A node was inserted into the page | 12 | – |
| Related style rule | 8 | – |
| A pseudo-class state changed (:hover, :focus, …) | 2 | – |
| Control | 3 | – |
| A stylesheet rule changed | 1 | – |
| An animation updated stylesAnimate transform and opacity so the compositor can run it off the main thread. | 1 | – |
| An element entered layout | – | 1 008 |
| An element left layout | – | 11 |
| Unknown | – | 2 |
| An element's size changed | – | 1 |
5 animations run on the main thread instead of on the browser's fast path that animates without blocking other page work (the compositor); they can make the rendering you see above janky. See them on the CPU tab.
The coach helps you find performance problems on your web page using web performance best practice rules. And gives you advice on privacy and best practices. Tested using Coach-core version 9.2.1.
decodingAsyncThe page has 1 image (out of 1) without a decoding hint. Add decoding="async" to non-critical images so the browser can decode them off the main thread.
Setting decoding="async" on an <img> tells the browser it can decode the image off the main thread, which keeps the page responsive to user interactions while images are being processed. The default ("auto") leaves the choice to the browser. https://developer.mozilla.org/en-US/docs/Web/HTML/Element/img#decoding
modernImageFormatsThe page ships 1 image (out of 1) in JPEG/PNG/GIF without a modern alternative. Wrap them in a <picture> with a <source type="image/avif"> or "image/webp" before the legacy <img>, or serve modern formats from your image pipeline directly. AVIF and WebP usually deliver 25–50% smaller files at the same quality.
AVIF and WebP routinely deliver 25–50% smaller files than JPEG and PNG at the same perceived quality, and every browser version still under support understands at least one of them. Ship modern formats either through a <picture> element with <source type="image/avif"> / "image/webp" entries in front of the legacy <img>, or directly from a content-negotiating image pipeline that returns AVIF / WebP when the client accepts it. https://web.dev/articles/serve-images-webp
longTasksThe page has 3 CPU long tasks with the total of 668 ms. The total blocking time is 59 ms and 2 long tasks before first contentful paint with total time of 559 ms. However the CPU Long Task is depending on the computer/phones actual CPU speed, so you should measure this on the same type of the device that your user is using. Use Geckoprofiler for Firefox or Chromes tracelog to debug your long tasks.
Long CPU tasks locks the thread. To the user this is commonly visible as a "locked up" page where the browser is unable to respond to user input; this is a major source of bad user experience on the web today. However the CPU Long Task is depending on the computer/phones actual CPU speed, so you should measure this on the same type of the device that your user is using. To debug you should use the Chrome timeline log and drag/drop it into devtools or use Firefox Geckoprofiler.
avoidRenderBlockingThe page has 1 blocking requests and 1 in body parser blocking (2 JavaScript and 0 CSS).
The critical rendering path is what the browser needs to do to start rendering the page. Every file requested inside of the head element will postpone the rendering of the page, because the browser need to do the request. Avoid loading JavaScript synchronously inside of the head (you should not need JavaScript to render the page), request files from the same domain as the main document (to avoid DNS lookups) and inline CSS for really fast rendering and a short rendering path.
avoidScalingImagesThe page has 1 image that are scaled more than 100 pixels. It would be better if those images are sent so the browser don't need to scale them.
It's easy to scale images in the browser and make sure they look good in different devices, however that is bad for performance! Scaling images in the browser takes extra CPU time and will hurt performance on mobile. And the user will download extra kilobytes (sometimes megabytes) of data that could be avoided. Don't do that, make sure you create multiple version of the same image server-side and serve the appropriate one.
cacheHeadersLongThe page has 4 requests that have a shorter cache time than one year (but still a cache time).
Setting a cache header is good. Setting a long cache header (a year) is even better because the asset will stay in the browser cache across visits. For content-hashed URLs (e.g. app.4af2.css) you can safely use Cache-Control: max-age=31536000, immutable. For unversioned URLs that may change, use a revalidating strategy instead.
viewportThe viewport meta tag does not contain width=device-width, the browser may use a desktop-width fallback.
The viewport meta tag tells the browser how to lay out the page on small screens. Without it (or without width=device-width) the page is rendered at a desktop fallback width and scaled down, which makes text unreadable on mobile. Disabling zoom (user-scalable=no, maximum-scale<=1) is also an accessibility regression. https://developer.mozilla.org/en-US/docs/Web/HTML/Viewport_meta_tag
unnecessaryHeadersThere are 5 responses that sets both a max-age and expires header. There are 6 responses that sets a server header.
Do not send headers that you don't need. We look for p3p, cache-control and max-age, pragma, server and x-frame-options headers. Have a look at Andrew Betts - Headers for Hackers talk as a guide https://www.youtube.com/watch?v=k92ZbrY815c or read https://www.fastly.com/blog/headers-we-dont-want.
referrerPolicyNo <meta name="referrer"> tag was found on the page. Set a Referrer-Policy response header (preferred) or add a meta tag, for example <meta name="referrer" content="strict-origin-when-cross-origin">.
Without an explicit referrer policy the browser falls back to the user-agent default and may leak the full URL of the previous page (including query strings) to every cross-origin request. Set a Referrer-Policy response header (preferred) or a <meta name="referrer"> tag in the document. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Referrer-Policy
contentSecurityPolicyHeaderSet a Content-Security-Policy header to mitigate cross-site scripting attacks. You can start with a Content-Security-Policy-Report-Only header, which only reports violations rather than blocking them.
A Content-Security-Policy response header tells the browser which sources of script, style, and other content are allowed. The most effective form is a strict CSP using nonces or hashes together with strict-dynamic; the worst is a missing header, with unsafe-inline and unsafe-eval close behind. https://web.dev/articles/strict-csp
crossOriginEmbedderPolicyHeaderSet a Cross-Origin-Embedder-Policy header (typically require-corp or credentialless) on the document response to control cross-origin embedding.
Cross-Origin-Embedder-Policy (COEP) makes the page refuse to load cross-origin subresources unless they explicitly opt in via CORP or CORS. Together with Cross-Origin-Opener-Policy it puts the page in a cross-origin isolated context, which mitigates cross-window side-channel attacks (Spectre) and unlocks high-resolution timers and SharedArrayBuffer. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Cross-Origin-Embedder-Policy
crossOriginOpenerPolicyHeaderSet a Cross-Origin-Opener-Policy header (typically same-origin) on the document response to isolate the page from cross-origin windows.
Cross-Origin-Opener-Policy (COOP) lets a page sever its window-group ties to cross-origin documents that opened it or that it opens. Together with Cross-Origin-Embedder-Policy it puts the page in a cross-origin isolated context, which mitigates cross-window side-channel attacks (Spectre) and unlocks high-resolution timers and SharedArrayBuffer. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Cross-Origin-Opener-Policy
crossOriginResourcePolicyHeaderSet a Cross-Origin-Resource-Policy header (same-origin, same-site or cross-origin) on the document response to limit who may embed it.
Cross-Origin-Resource-Policy (CORP) is a per-response opt-in that tells the browser which origins are allowed to embed the resource. It blocks cross-origin or cross-site no-cors embedding (img, script, iframe, etc.) and is one of the building blocks of cross-origin isolation. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Cross-Origin-Resource-Policy
permissionsPolicyHeaderSet a Permissions-Policy header to control which browser features the page can use.
The Permissions-Policy response header (the successor to Feature-Policy) lets a site explicitly opt in or out of powerful browser features such as camera, microphone, geolocation, payment and clipboard. Setting a strict policy reduces the attack surface and limits what embedded third parties can do. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/Permissions-Policy
referrerPolicyHeaderSet a referrer-policy header to make sure you do not leak user information.
Referrer Policy is a new header that allows a site to control how much information the browser includes with navigations away from a document and should be set by all sites. https://scotthelme.co.uk/a-new-security-header-referrer-policy/.
xContentTypeOptionsHeaderSet X-Content-Type-Options: nosniff on the document response to prevent MIME-sniffing.
X-Content-Type-Options: nosniff prevents browsers from interpreting files as a different MIME type than what is declared in the Content-Type header. This blocks a class of cross-site scripting and content-type confusion attacks and should be set on every response. https://developer.mozilla.org/en-US/docs/Web/HTTP/Headers/X-Content-Type-Options
A snapshot of what the browser actually built for this page: the document, how big and deep the DOM tree is and what it kept in storage. Big, deep trees are slower to style and lay out, so the counts Chrome warns about carry a flag.
Document
What the page says it is and how large it rendered.
DOM structure
How much markup the browser has to build, style and lay out.
What the page stored on the client, and the network it was tested on.
Data collected using
Coach-core version 9.2.1. With updated code from
Webappanalyzer 2026-05-04. Use
--browsertime.firefox.includeResponseBodies html or
--browsertime.chrome.includeResponseBodies html to help Wappalyzer find more information about technologies used.
Data from run 2
0.0s0.6s1.1s1.2s1.3s1.4s1.6s1.7sWhen the page main content is rendered, collected via the Largest Contentful Paint API. Read more about Largest Contentful Paint.
body#www-wikipedia-org > main > div:eq(0) > h1 > span
The LCP element is highlighted in the screenshot. If nothing is highlighted the element was removed before the screenshot or the LCP API couldn't find it.
The Largest Contentful Paint API matched this image:
How much the page's content shifts as it loads, collected via the Cumulative Layout Shift API.
One shift is 100% of your 0.001 CLS. div:eq(0) moved at 1.588 s. Reserve space for whatever renders into that spot, a fixed height or an aspect-ratio box, so later content stops pushing the page around.
Worst first, showing shifts of 0.01 and up. Each percentage is that shift's exact share of the score.
The shift is spread across many movements below 0.01, none large enough to single out. Check the video or filmstrip to see what moved.
Elements that shifted by more than 0.01 are highlighted in the screenshot. If an element shifted outside the viewport, it won't appear here. Check the video or filmstrip to see the shift.
Timing data the server chose to expose through Server-Timing response headers on the main document — typically backend time, cache status or experiment flags.
| Name | Duration | Description |
|---|---|---|
cache | 0 ms | hit-front |
host | 0 ms | cp3070 |
co_id | 0 ms | 1382805900 |
There are no custom configured scripts.
There are no custom extra metrics from scripting.
How the page is built.
| Header | Value |
|---|---|
accept-ranges | bytes |
age | 52029 |
cache-control | s-maxage=86400, must-revalidate, max-age=3600 |
content-encoding | gzip |
content-length | 30635 |
content-type | text/html |
date | Mon, 27 Jul 2026 09:27:36 GMT |
etag | W/"1d629-65738a87fd940" |
last-modified | Wed, 22 Jul 2026 20:05:17 GMT |
nel | { "report_to": "wm_nel", "max_age": 604800, "failure_fraction": 0.05, "success_fraction": 0.0} |
report-to | { "group": "wm_nel", "max_age": 604800, "endpoints": [{ "url": "https://intake-logging.wikimedia.org/v1/events?stream=w3c.reportingapi.network_error&schema_uri=/w3c/reportingapi/network_error/1.0.0" }] } |
server | ATS/9.2.13 |
server-timing | cache;desc="hit-front", host;desc="cp3070",co_id;desc="1382805900" |
strict-transport-security | max-age=106384710; includeSubDomains; preload |
x-analytics | |
x-cache | cp3070 miss, cp3070 hit/3271033 |
x-cache-status | hit-front |
x-client-ip | 138.201.135.103 |
x-request-id | 5a1e08e1-9938-4efe-b0a8-e27804a4835c |
| URL | Type | Transfer Size | Content Size |
|---|---|---|---|
| https://www.wikipedi...ia-logo-v2@2x.png | image | 37.6 KB | 36.6 KB |
| https://www.wikipedia.org/ | html | 31.2 KB | 117.5 KB |
| https://www.wikipedi...rite-e49fbf32.svg | svg | 18.8 KB | 50.0 KB |
| https://www.wikipedi...dex-34f340e24a.js | javascript | 9.2 KB | 23.6 KB |
| https://www.wikipedi...con/wikipedia.ico | favicon | 1.8 KB | 2.7 KB |
| https://www.wikipedi...ie9-507b16b6be.js | javascript | 1.4 KB | 580 B |
Transfer size100.1 KB
Content size230.9 KB
Requests6
| Content | Header Size | Transfer Size | Content Size | Requests |
|---|---|---|---|---|
| html | 0 b | 31.2 KB | 117.5 KB | 1 |
| javascript | 0 b | 10.6 KB | 24.1 KB | 2 |
| image | 0 b | 37.6 KB | 36.6 KB | 1 |
| favicon | 0 b | 1.8 KB | 2.7 KB | 1 |
| svg | 0 b | 18.8 KB | 50.0 KB | 1 |
| Total | 0 b | 100.1 KB | 230.9 KB | 6 |
| Domain | Total download time | Transfer Size | Content Size | Requests |
|---|---|---|---|---|
| www.wikipedia.org | 1.254 s | 100.1 KB | 230.9 KB | 6 |
| type | min | median | max |
|---|---|---|---|
| Expires | 1 hour | 1 day | 1 year |
| Last modified | 4 days | 4 days | 3 weeks |
Includes requests done after load event end.
| Content | Transfer Size | Requests |
|---|---|---|
| html | 0 b | 0 |
| css | 0 b | 0 |
| javascript | 0 b | 0 |
| image | 0 b | 0 |
| font | 0 b | 0 |
| favicon | 1.8 KB | 1 |
| Total | 1.8 KB | 1 |
Includes requests done after DOM content loaded.
| Content | Transfer Size | Requests |
|---|---|---|
| html | 0 b | 0 |
| css | 0 b | 0 |
| javascript | 0 b | 0 |
| image | 0 b | 0 |
| font | 0 b | 0 |
| favicon | 1.8 KB | 1 |
| Total | 1.8 KB | 1 |
JavaScript and CSS coverage from the extra profile run made by --enableProfileRun, not from the timed iterations. For JS, Chrome tracks which blocks of each script actually ran: a byte is unused when the innermost block of code around it never ran. For CSS, Chrome records every selector that no element on the page matched during the iteration. Enable with --chrome.coverage on every run, or --enableProfileRun to keep the main run's timing metrics untouched. Unused bytes still cost network, parse and compile time for no benefit. Expand the section below for the full picture.
sitespeed.io drives Chrome's V8 profiler and CSS rule tracker via the DevTools Protocol, the same machinery the DevTools Coverage panel uses. The two columns are populated by two different mechanisms.
Browsertime calls Profiler.startPreciseCoverage with detailed (block-level) mode at the start of the iteration and Profiler.takePreciseCoverage at the end. V8 returns nested ranges per function: an outer range covering the function body and inner ranges for sub-blocks (if/else, switch arms, ternaries). Each range has an execution count. A byte is "used" if the innermost range that covers it has count > 0, "unused" if it has count = 0, so we catch both functions that never ran AND dead branches inside functions that did.
Browsertime calls CSS.startRuleUsageTracking at the start of the iteration and CSS.stopRuleUsageTracking at the end. Chrome returns a flat list of rule-text ranges with a used flag: a rule is "used" if at least one element on the page matched it during the iteration. Both inline <style> blocks and external stylesheets are included.
Coverage reflects only what executed during this run. Code behind a button click stays unused unless the iteration clicked the button. The collection window opens before navigation and closes when measure.stop() is reached, so the only way to count interactions is to drive them between measure.start() and measure.stop() inside a navigation script (pass a .js file as the input instead of a plain URL). Clicks, form fills and waits placed there all count toward "used". --preScript runs before the URL is loaded so it can't interact with the page being measured, and --postScript runs after coverage has already been collected, so neither helps here. Detailed coverage also tells V8 to skip some optimisations, which drifts timing metrics like LCP and TBT; that's why --enableProfileRun is the safer option for production measurement (it collects coverage in a separate iteration that doesn't count towards the medians).
84.7 KB of script and style shipped, and 43.5 KB of it never ran on this page. The biggest win is www.wikipedia.org: 59.0 KB delivered, 28.3 KB unused.
| URL | Total | Used | Unused | Unused % |
|---|---|---|---|---|
| https://www.wikipedia.org/portal/wikipedia.org/assets/js/index-34f340e24a.js | 23.5 KB | 8.4 KB | 15.1 KB | |
| https://www.wikipedia.org/portal/wikipedia.org/assets/js/gt-ie9-507b16b6be.js | 580 B | 513 B | 67 B | |
| https://www.wikipedia.org/ 2 scripts | 1.6 KB | 1.6 KB | 0 b |
| URL | Total | Used | Unused | Unused % |
|---|---|---|---|---|
| https://www.wikipedia.org/ 3 stylesheets | 59.0 KB | 30.6 KB | 28.3 KB |
A high unused-% on one run means this iteration loaded code that didn't get exercised, typical on first-load journeys before any user interaction. Drive the real flow by passing a navigation script as the input instead of a plain URL, with clicks and form fills placed between measure.start() and measure.stop(); anything in that window counts toward "used". The advice below is for code that's still unused after a representative run.
import(). Bundlers turn the import boundary into a separate chunk.defer or async so the parser doesn't block on code the page doesn't need yet.media= tricks or rel="preload" swaps.Tasks ≥ 50 ms blocking the main thread, collected via the Long Task API.
A long animation frame (LOAF) is a frame that took ≥ 50 ms from input to the next paint. The part beyond the threshold is blocking time, the number that hurts Interaction to Next Paint and Total Blocking Time. Break long tasks up (scheduler.yield(), smaller chunks) or move the work off the main thread. Read more about the Long Animation Frames API.
| Script | Frames | Time running | Blocking (share by run time) | Forced style & layout | Triggered by |
|---|---|---|---|---|---|
| www.wikipedia.org | 1 | 7.9 ms | 3.4 ms | – | script tag |
A frame's blocking time is split between its scripts by their share of the frame's work, capped at each script's own run time — blocking from work no script owns (HTML parsing, browser internals) stays unattributed. The per-script blocking view below uses the same attribution.
The browser did not report which script was responsible for this frame — typically its own style/layout work, or a cross-origin script it will not name.
The browser did not report which script was responsible for this frame — typically its own style/layout work, or a cross-origin script it will not name.
2 more long frames carried no blocking time (nothing ran past the 50 ms threshold) — long, but harmless to INP and TBT.
How much each script blocked the main thread, derived from the Long Animation Frame API. Each frame's blocking time is split between its scripts by their share of the frame's work, capped at each script's own run time — blocking from work no script owns (HTML parsing, browser internals) is not attributed (older data credits the script that started the frame). The closest answer to "which script should I fix to improve TBT" the platform exposes.
Total Blocking Time attributed per script from the Chrome trace: every main-thread task longer than 50 ms blocks input for the remainder, and each blocking task is credited to the script whose own code ran the longest during that task.
All main-thread work during the full page load, not just the share that blocked input. Calculated from the Chrome trace.
parseHTMLstyleLayoutscriptParseCompilescriptEvaluationpaintCompositeRendergarbageCollectionotherThe functions where main-thread JavaScript time was actually spent, sampled by Chrome's JavaScript profiler. Self time is spent in the function itself; total time includes the functions it called. For loader/dispatcher functions whose total dwarfs their self time, the runs line shows the top functions and modules (up to five, ≥ 1 ms) the total was actually spent running.
How much setTimeout/setInterval work the page runs, credited to the script that scheduled each timer. Many zero-delay timers mean work sliced into a machine-gun of tiny tasks; repeating timers are polling.
| Set by | Delay | Type | Times set | Times fired | Fire time |
|---|---|---|---|---|---|
| https://www.wikipedia.org/:941 | 1000 ms | once | 1 | 1 | 0.0 ms |
Main-thread CPU time from the Chrome trace split by domain. First party is wikipedia.org.
Animations that fell back from the compositor to the main thread, blocking each frame instead of running on the GPU.
Each chip below is a CSS property the page tried to animate that Chrome couldn't hand to the compositor. Swap it for a transform or opacity equivalent where you can.
border-bottom-color1×border-left-color1×box-shadow1×border-right-color1×border-top-color1×Want to dig deeper? Download the Chrome trace and drag-and-drop it into Performance in DevTools.