The rules of the game
web.dev states the good thresholds as LCP within 2.5 seconds, INP of 200 milliseconds or less and CLS of 0.1 or less, measured at the 75th percentile of page loads and segmented across mobile and desktop. developer.chrome.com states that the Chrome UX Report is a 28-day rolling aggregate of opted-in Chrome users' experiences, that its API reports the p75, and that page-level data has stricter eligibility than origin-level data. INP became a stable Core Web Vital on 12 March 2024, replacing First Input Delay. Adobe's documentation adds that on Adobe Commerce on cloud infrastructure, Fastly is included and configured through custom VCL snippets with per-environment credentials, so edge behavior is the merchant's or the agency's to change.
LCP: what is published
ParadoxLabs publishes the most precise LCP figures on the page and labels them: Denver Wholesale Foods from 4.1 to 1.4 seconds mobile p75 LCP in 2026, an Adobe Commerce to Mage-OS migration from 1.7 to 1.3 seconds mobile p75 LCP, and Half Time Beverage's share of good LCP experiences from 52% to 88% in the month the work shipped, sourced to CrUX for July to August 2023. scandiweb's Läderach case gives homepage LCP of 1.8 seconds on desktop and 1.9 on mobile six months after the first meeting, its Classic Football Shirts case reports a 15% LCP improvement, and its Scouting America case reports LCP improved by 12 to 13% on mobile and desktop; none of the three names CrUX or a window. scandiweb's 2026 performance guide puts stock Magento at a median LCP near 3.8 seconds, Luma at roughly 4.8, and Hyvä sites at an average near 1.2 seconds. Its 2023 roundup states that on Nicokick 73% of all pages had LCP at or under 2.5 seconds after the work. elgentos names LCP among the metrics its Varnish reconfiguration improved on Haibu, without a value. MGT-Commerce, Sonassi, integer_net, Snowdog and Bemeir publish no LCP figure on a client.
INP: what is published
scandiweb's Classic Football Shirts case reports a 45% INP improvement after a Hyvä migration paired with an upgrade from 2.4.0 to 2.4.4-p2, and its performance guide states that INP responds to lighter JavaScript, which is what bundling and Hyvä deliver, with an audit sample where desktop total blocking time fell from 2,290 to 47. Its Läderach case reports the older FID metric at 3 milliseconds on desktop and 20 on mobile. elgentos's Haibu write-up names cart interaction work for INP as one of the changes behind its p90 result. Bemeir's article on express checkout wallets is about keeping payment SDKs from hurting Core Web Vitals, without a measurement. The rest publish no INP figure. The cause is well documented across the page: Luma's RequireJS and Knockout stack, and the fix is a lighter frontend, which is why Hyvä appears in every INP conversation.
CLS: what is published
Almost nothing with a number. scandiweb's Läderach case states the store hit the CLS target for most users, and its Scouting America case states improvements in CLS and FCP, without values. No other agency on the page publishes a CLS figure. web.dev's threshold of 0.1 is the number to write into the exit criterion, and Search Console's report by URL group is where a merchant will find the current value, because none of the eight publishes one in a case study on the pages read.
Hyvä: what is published
The theme itself is free and open source from 10 November 2025 under OSL 3.0 and AFL 3.0, per hyva.io's announcement, with Checkout, Enterprise and Commerce remaining commercial. What the agencies publish about its effect falls into three kinds. General claims come first. integer_net's Hyvä page states a Lighthouse score of over 90 by default even on mobile; Snowdog's Hyvä page states a perfect 100/100 PageSpeed score; MGT-Commerce's guide states a product page of 0.15 MB against 0.9 MB on Luma and a throttled mobile product page at 1.6 seconds against 8.0; scandiweb's guide states a Core Web Vitals score around 88 against 31 on Luma. Client cases come second. scandiweb's Byggmax, where category page PageSpeed went from 85 to 99 and product pages from 70 to 87 after moving only those templates and leaving the checkout, and its Läderach, Classic Football Shirts and Scouting America cases above. Technical writing is the third kind. Bemeir's article on speculation rules, which Hyvä ships from version 1.4 defaulted to prefetch with four eagerness settings, and elgentos's free PDF on Hyvä frontend development. hyva.io itself claims Hyvä Checkout loads 13 times faster on mobile than the standard Magento checkout, which is the vendor's claim.
Measurement: field, lab and the percentile argument
Every PageSpeed and Lighthouse figure above is lab data: one throttled run on one URL. Every CrUX figure is field data at p75 over 28 days. elgentos makes the case in its Haibu post that p90 is the number to watch, because a store that is fast for its slowest 10% of visitors is fast for everyone, and it publishes a Magento 2 module for RUMvision, the real user monitoring tool it used, on GitHub. scandiweb's audit page promises a Core Web Vitals scorecard for key pages, which is a field view by URL group. The gap on this page is not that agencies lack results; it is that only one labels them. A merchant can close the gap in the contract by writing the exit criterion as p75 field data by URL group over a full 28-day window, and by asking every agency on the shortlist for the Search Console view of a comparable store before the first call.
| Metric | Good threshold (web.dev) | Agencies with a client figure on their own pages | Field-labeled |
|---|---|---|---|
| LCP | 2.5 seconds or less at p75 | ParadoxLabs (three cases), scandiweb (Läderach, Classic Football Shirts, Scouting America, Nicokick) | ParadoxLabs only |
| INP | 200 milliseconds or less at p75 | scandiweb (Classic Football Shirts) | None labeled |
| CLS | 0.1 or less at p75 | None with a value | None |