This page compares the five performance numbers you paste with the thresholds Google publishes on web.dev. LCP, INP and CLS are the three Core Web Vitals; TTFB and FCP are supporting metrics that often explain why a core vital is slow. Each value is rated Good, Needs improvement or Poor, and the distance to the nearest threshold is shown.
It does not fetch your page, call the PageSpeed Insights API or read field data. You bring the numbers from a report, and the page does the comparison and the arithmetic in the browser tab.
The five metrics and their thresholds
LCP, Largest Contentful Paint, is when the main image or heading finishes rendering. Good is 2,500 ms or less; above 4,000 ms is poor. INP, Interaction to Next Paint, is the delay between a tap or keypress and the next frame. Good is 200 ms or less; above 500 ms is poor. INP replaced First Input Delay as a Core Web Vital in 2024.
CLS, Cumulative Layout Shift, measures unexpected movement. Good is 0.1 or less; above 0.25 is poor. TTFB, Time to First Byte, is the server's first response: good is 800 ms or less, poor above 1,800 ms. FCP, First Contentful Paint, is when the first text or image appears: good is 1,800 ms or less, poor above 3,000 ms.
The boundaries are inclusive. A value of exactly 2,500 ms, 200 ms or 0.1 is rated Good, which is why a page sitting on the line still passes. Google's published thresholds are assessed at the 75th percentile of page loads, separately for mobile and desktop, so a single fast load does not make a page pass.
A worked example
Five values were entered with the source set to mixed and the device to mobile: LCP 4,200 ms, INP 350 ms, CLS 0.30, TTFB 900 ms and FCP 3,100 ms. The gauge read 0 out of 100 and the band was Failing.
| Metric | Value | Good up to | Poor above | Rating |
|---|---|---|---|---|
| LCP | 4,200 ms | 2,500 ms | 4,000 ms | Poor |
| INP | 350 ms | 200 ms | 500 ms | Needs improvement |
| CLS | 0.30 | 0.1 | 0.25 | Poor |
| TTFB | 900 ms | 800 ms | 1,800 ms | Needs improvement |
| FCP | 3,100 ms | 1,800 ms | 3,000 ms | Poor |
All five rows were outside the good band, so the score was zero. The status line was narrower: it said 0 of 3 Core Web Vitals are in the good band, because the gauge divides by every metric entered while the headline counts only LCP, INP and CLS. With all three core vitals failing, both readings agree. With a mix of values they can disagree, and the gauge is the one to watch if you care about supporting metrics too.
Because every row failed, the page also printed a fix for each one: find the LCP element and the delay, split long JavaScript tasks for INP, reserve space for CLS, check hosting and caching for TTFB, and remove render-blocking work for FCP. The advice is generic on purpose; the tool cannot see which element or script is responsible.
Two places where the wording misleads
The worst metric label is not reliable when everything passes. The page sorts by rating first, then by headroom, and for a good metric headroom is the distance to the good threshold. The largest headroom wins the sort, so a page with every metric passing can name its worst metric as the one with the most room to spare. In the all-good test, FCP was named as the worst metric with 200 ms of headroom while LCP had 100 ms.
The improvement figure is also measured to the wrong line for a metric in the middle band. For a value between the two thresholds, the page calculates the distance to the poor threshold, then calls it the improvement needed to reach the good band. A TTFB of 900 ms was told it needed 900 ms of improvement to reach 800 ms; the real gap is 100 ms. With INP at 350 ms the two distances happen to be equal, so that row reads correctly by coincidence.
Neither problem changes the rating. They change the sentence beside it, so read the table when the wording looks odd, and subtract the value from the good threshold yourself if you need the true gap.
Field data, lab data and the 75th percentile
The source control changes the warning, not the maths. Choose field for numbers from the Search Console Core Web Vitals report or the CrUX dashboard, lab for a single Lighthouse or PageSpeed Insights run, and mixed only when you are deliberately comparing the two. A lab score is one synthetic load; field data is what real visitors experienced.
Mixing the two produces a number that means nothing in particular, and the page says so. It also reminds you that Search Console needs a minimum amount of field data before it will assess a URL group. A new page can be missing from the report for weeks, and that absence is not a failure.
If you are checking a page you have just changed, compare lab runs before and after with the same device and the same network settings. For a pass or fail judgement that matches Search Console, use field data over 28 days and look at the 75th percentile rather than the average.
What to fix first
Start with poor ratings, then the largest gap. LCP and TTFB are often connected: a slow first byte delays everything after it, so check hosting, caching and redirects before compressing images. If TTFB is good and LCP is not, the delay is usually in the hero image, a render-blocking stylesheet or a late font.
For INP, look for long tasks on click and keypress rather than the total JavaScript bundle. For CLS, set width and height on images and embeds, and avoid inserting banners above existing content. These are the same priorities the tool lists, and none of them need a paid API.
The log file analyser is a useful companion when TTFB is the problem: a slow server response shows up in access logs as time on the request, and a 5xx burst explains a bad run that a lab test cannot reproduce.