Faragiri

Learn Core Web Vitals Speed

A Devon farm shop failed on one measurement of three, and the number that failed was the one nobody in the business had ever heard a name for.

4Sections
3Min read
4Depth

Three measurements decide the assessment and a page passes only when all three pass. Ottery Organics runs a farm shop and an online box scheme in Devon, and in May 2026 its shop pages measured 1.9 seconds, 140 milliseconds and 0.31. Two of those are comfortable and the third fails, and the third is the one nobody in the business had ever heard a name for.

LCP under 2.5 seconds is when the biggest thing on screen finishes drawing. INP under 200 milliseconds is how long the page takes to respond visibly after somebody touches it. CLS under 0.1 is how far things move after they have appeared. Those three sentences are the whole vocabulary and they are worth learning verbatim, because every report about this subject assumes them.

The one with no unit

CLS has no unit, because it multiplies distance moved by area affected. That is why 0.31 is not a distance, a time or a percentage of anything: a banner that shifts the whole page by a little scores like a small element that jumps a long way, which is deliberate and is the reason the number is unintuitive.

Core Web Vitals

The three measurements at one farm shop in May 2026

They do not compare with each other: one is a time, one is a shorter time, and one is a ratio with no unit at all.

  • 1.9 secondsLargest Contentful PaintThe vegetable box photograph. Threshold is 2.5 seconds and this passes.
  • 140 millisecondsInteraction to Next PaintThe worst interaction in a visit, not the first. Threshold is 200 milliseconds and this passes.
  • 0.31Cumulative Layout ShiftDistance moved times area affected, pushing text down about 180 pixels. Threshold is 0.1 and this fails.

Adding width and height attributes to the box photograph stops the 180-pixel shift so all three pass.

The farm shop failed on CLS at 0.31 and passed the other two, and the cause was an image with no width and height on it. The vegetable box photograph loaded after the text, the text moved down by about 180 pixels when it arrived, and anybody who had started reading lost their place. The fix is two attributes on an image tag.

What the other two are measuring

LCP is usually an image or a heading and it is worth knowing which. At Ottery it is the box photograph, so the same file that caused the layout shift is also the thing the measurement is waiting for, and one change moved both.

LCP and INP

What the two timed measurements actually wait for

One times how fast the main visual arrives; the other times the slowest response across an entire visit.

Largest Contentful PaintVisual loading milestone.
Interaction to Next PaintReplaced FID in March 2024.
What it measuresWhen the biggest image or heading finishes drawingTime to visible response after a touch or click
Which moment countsFinal render of the main above-the-fold elementWorst interaction in the whole visit, not the first
Why prior passes breakSame target element as beforePages that passed FID can fail INP with unchanged code
At Ottery OrganicsSame vegetable box photograph that caused the shiftWell under the 200-millisecond threshold

Fixing the box photograph file affected both CLS and LCP because the same element controlled both.

INP replaced FID in March 2024 and measures the worst interaction, not the first. That change matters for anything written before it: a page that passed the old measurement comfortably can fail the new one without a line of its code changing, because the old one timed the first response and this one takes the worst across a whole visit.

Where this stalls: measuring one page

The assessment covers a page group, not one page, when data is thin. Ottery has 140 product pages and only the shop index has enough visitors to be measured on its own; the rest are assessed as a group, so a fix to one product page shows up nowhere and a fault on the template shows up everywhere.

Measurement unit

How Google assesses thin pages vs busy pages

Optimising a single product page moves nothing when traffic is only large enough to assess the template group.

Does the single URL have enough visitors to measure alone?
  • Yes (shop index)
    Measured as an individual pageTraffic is heavy enough for its own field score.
    • Changes on that URL appear directly in its report
  • No (140 product pages)
    Grouped by shared templateThin pages are pooled together into one page group.
    • Editing one product page shows up nowhere
    • Fixing the shared template lifts all 140 pages

Work at the template level and read results at the group level rather than watching one URL.

The stall is a fortnight spent optimising a single page and watching a number that was never about that page. The useful unit of work here is the template, and the useful unit of measurement is the group, and neither of those is what a tool asks you to type into a box.

What to open, and in what order

PageSpeed Insights shows the field assessment first and a lab run second, free, and it replaced running a lab tool alone and assuming its score was the number being used. Read the top half. The bottom half is a diagnosis and it answers a different question, which is why your score and your report disagree.

Workflow

What to open, and in what order

Start with real field status and site-wide page groups before touching lab simulations.

  1. 01Open PageSpeed Insights and read the top halfThe top panel holds the field assessment; the bottom lab score answers a different diagnostic question.
  2. 02Ignore lab vs field score disagreementsDo not mistake a simulated lab score for the pass or fail status used in search.
  3. 03Check Search Console's Core Web Vitals reportInspect every page group at once instead of testing individual URLs and hoping they match.
  4. 04Fix the shared template faultAt Ottery, one fault on the vegetable box photograph explained 140 failing product pages.

Reading Search Console's group report first stops teams from treating one template bug as 140 separate jobs.

Search Console has a Core Web Vitals report covering every page group on the site at once, free, and it replaced checking pages one at a time and hoping they were representative. It is the only place the group structure is visible, and at Ottery it showed one group of 140 pages failing on one measurement, which is a single template fault stated as a hundred and forty failures.

What each number is worth once it passes is the honest size of the ranking effect, and the effect on how much of a large site gets fetched at all is what a slow site costs you in pages indexed. Both sit under how fast it loads, and what that is worth in search, and the vegetable box photograph is the whole of Ottery's version of it.

Alongside this

Above this