Performance is the frontend topic where opinion is cheapest and data is most available, so it is where a candidate can differentiate fastest simply by knowing the actual field numbers.

All three are assessed at the 75th percentile of page loads, segmented between mobile and desktop. That percentile choice is deliberate and worth saying out loud: it means a quarter of your users can have a worse experience than your reported number.

Interaction to Next Paint fully replaced First Input Delay — stable in March 2024, with the old metric removed from tooling that September. The change mattered because the old one measured only the delay before the first interaction was processed, so a page could score perfectly while every interaction after the first was slow. The new one measures the whole interaction through to the next paint, across the whole visit. There is still no fourth Core Web Vital as of August 2026.

One genuinely important thing in flight: the Soft Navigations API, which heuristically detects single-page-application route changes so the metrics can attribute them. Final origin trial in Chrome 147, unflagged from Chrome 151. This matters enormously, because today a single-page app gets one paint measured at the hard navigation and nothing thereafter — every subsequent route change is invisible to field data.

From the Chrome user-experience dataset, July 2026: just over 18 million origins, of which 55.7% pass all three Core Web Vitals — up from about 51% eighteen months earlier. So nearly half the web still fails, and the trend is improvement of roughly three points a year.

From the 2025 Web Almanac, which is the richest measured source available: mobile 48% passing, desktop 56%. Individually, interaction latency passes on 97% of desktop pages and 77% of mobile ones; loading passes on 74% and 62%.

Some context behind those rows. The largest element is an image on about 85% of desktop pages and 76% of mobile ones — so the loading metric is, in practice, an image delivery metric. And image formats are barely modernised: JPEG is still 57% of largest-element images, PNG 26%, and WebP only 11%.

On layout stability: 87% of mobile pages load a web font but only 22% pre-connect to its origin and about 15% preload it, and 40% of mobile pages run animations that are not composited — meaning they animate properties that force layout rather than running on the compositor.

Two categories have named case studies with business metrics attached, which is much stronger evidence than the usual advice:

• Speculation Rules — prerendering the likely next navigation. One eyewear retailer published mobile loading dropping from 4.69 to 2.66 seconds with mobile conversion up 101% and desktop up 156%; a publisher reported loading down 17.9% with engagement peaks up 8.9%. Say the caveat out loud though, because it is commonly misunderstood: this improves the next page, not the current one. Adoption is only 3–5% among the largest sites but around 24% in the long tail, almost entirely because one content management system ships it by default.
• Interaction latency work, with published outcomes: a property platform cut it 80% and reported 36% more conversions year over year; a travel site cut it 72% for a 7% sales lift; a news publisher cut blocking time thirtyfold and halved its bounce rate; a retailer cut it 50% for a click-through gain.

The mechanics behind that second group are the same everywhere: break long tasks so the browser can paint between them, using the scheduler yield API, and attribute the remaining ones with the Long Animation Frames API rather than guessing. Note that a widely repeated ‘40–60% improvement from yielding’ figure could not be traced to a primary source — use the named case studies instead.