Verbatim text from the 100-item checklist; this site does not re-judge it. The measured values on the right are from a single fetch and are not paired line by line.
Fields come from this tool's M0 detector (a single fetch), with no day-over-day change or verdict; vantage server-KR. The checklist criteria on the left are verbatim and are not re-judged here.
Full report
Grouped by checklist item: criteria, general explanation, recommendations and official references are listed per item; the full set of measured values is listed separately at the end, not paired item by item. This site does not re-judge, and the explanations and recommendations do not give a compliance verdict on any single measurement.
Tool 性能与体验 · Site mvmt.com · Vantage server-KR · Generated 2026-11-22T21:06:05.534Z · Reference version 2026-11-09 · 6 checklist items
Generated on this page from checklist references and this run's measurements; no history, change events or compliance verdicts.
Checklist items · 6
#9 CDN 与全球响应耗时 · P1
Criteria:TTFB、边缘节点命中率、各区域下载速度
Explanation
Users and agents of a cross-border store sit in different regions, and TTFB decides whether a fetch completes before the timeout. Edge misses, cross-continent origin fetches and uncacheable dynamic pages make the same site look completely different from different vantages. Absolute values from different vantages are not comparable; follow the trend from one vantage.
Recommendations
Serve product pages and static assets from CDN edges with explicit cache headers and cache keys.
Cache mostly-static HTML at the edge and isolate personalised fragments via edge includes or separate requests.
Choose edge locations near your main markets and record TTFB trends per vantage.
Do not conclude the site is fast globally from one number; keep a separate baseline per vantage.
LCP, INP and CLS are Google's field metrics for real user experience and also shape whether a shopping agent treats a page as usable. Lab scores and field distributions often disagree, and a single reading is noisy; look at monthly field percentiles from CrUX-style data rather than one Lighthouse score.
Recommendations
Use CrUX / Search Console for mobile field distributions and watch percentile changes weekly.
Make the LCP resource discoverable and prioritised; reserve dimensions for images and ad slots to cut CLS.
Break up long tasks and reduce main-thread blocking to improve INP.
Allow at least a week after a change before judging impact; do not treat a same-day score as a verdict.
Most cross-border traffic and shopping agents use a mobile UA. Horizontal scroll, tiny tap targets, forced desktop redirects and undersized text keep an agent from parsing the main content or make it judge the page unusable. Typical failures: missing viewport, buttons far below 44px, no foldable adaptation, or a UA redirect to a stripped-down mobile subdomain.
Recommendations
Declare a correct viewport and keep primary buttons and links at a usable tap-target size.
Fetch the homepage and a product page with a mobile UA and compare main content against desktop.
Avoid UA redirects to a stripped-down mobile subdomain; never hide key content behind hover.
Regression-test the add-to-cart to checkout path on both narrow and foldable screens.
Images dominate page weight and are also how an agent judges whether product information is complete. Oversized originals, missing alt text and first-screen images injected by lazy-load scripts produce a page people can see but agents cannot read. Typical failures: WebP with no fallback, empty alt, or the hero image absent from the initial HTML.
Recommendations
Serve WebP/AVIF with a fallback and generate multi-size srcset matching container widths.
Give every product image a descriptive alt, never a filename or an empty string.
Put the first-screen hero image directly in the HTML rather than injecting it via lazy-load.
Compress before upload and periodically sample the site-wide image size distribution.
Many external agents do not execute JavaScript, or execute only part of it. If first-screen content depends on client-side rendering, the agent sees an empty shell; render-blocking CSS/JS also delays readable content. Typical failures: product data filled in only by hydration, uncompressed/uncached bundles, or critical styles loaded asynchronously.
Recommendations
Put product name, price, availability and structured data in server-rendered HTML or JSON-LD.
Compress and cache static assets, inline critical CSS and defer third-party scripts.
Fetch once with JavaScript disabled to confirm main content and prices are still readable.
Keep analytics, chat and A/B scripts from blocking first-screen rendering.
Contrast, missing labels and keyboard reachability are both a compliance requirement in Europe and the US and a direct input to how an agent understands form and control semantics. Unlabelled inputs, icon-only buttons and missing landmarks stop automated flows from finding add-to-cart or checkout. Typical failures: divs used as buttons, icon buttons with no accessible name, or forms with only a placeholder.
Recommendations
Give every form control a real label; use native button/a elements with accessible names.
Keep add-to-cart through checkout keyboard-reachable and never bind key actions to hover alone.
Scan key templates regularly with axe / Lighthouse and check contrast against WCAG AA.
Bake accessibility checks into the template release process instead of spot-checking after launch.