Core Web Vitals Repair Without Cutting Header Bidding

The vital that actually breaks: On ad-funded article pages the failing metric is almost always INP, not LCP. In the field traces we have taken on publisher templates, the ad stack contributes between 300 and 900 milliseconds of main-thread blocking inside the first five seconds, with the consent script, the ad library, the bidder adapters and analytics queued behind one another. A tap landing in that window is what puts the page past 200 milliseconds at the 75th percentile. The repair is scheduling, not deletion. How we find it: Field data or nothing. We instrument with the web-vitals attribution build and read long animation frames, so an INP failure comes back naming the script that blocked, the interaction target and the phase: input delay, processing, or presentation. Lab runs are useful for confirming a change and useless for finding the cause, because your worst interactions happen on a mid-range Android on a warm page that no lab run reproduces. We take CrUX for the 75th percentile you are judged on and our own instrumentation for the attribution. The repairs, in the order they pay: The consent stub goes inline and small, the rest of the consent script goes async, and neither sits in front of the largest image. Bidder initialisation is broken across yields instead of running as one block. The auction timeout is set against your own measured bidder response distribution rather than the round number someone inherited. Analytics and social embeds move behind an intersection or an interaction. Below-fold iframes get lazy loading. For layout shift, every slot gets a reserved box per breakpoint that matches the size mapping in the ad server; most publisher CLS is a slot allowed to accept a creative size the layout never budgeted for. For the largest paint, the hero gets fetch priority and nothing from the ad stack is allowed above it in document order. The trade-off we insist on measuring: Some of these changes cost money. A shorter auction timeout drops slow bidders, and slow bidders sometimes pay the most. Every change runs against a revenue holdout and you get the number, so trading milliseconds against a point of yield is your decision made against evidence rather than ours made quietly. Anyone telling you a speed change is free has not measured one. What this is not: This is not an SEO purchase. Page experience is not a ranking system, and if the problem is that a competitor outranks you, this work will not fix it and we will say so before you buy rather than after. Buy it when your readers are on slow devices, when your subscription funnel loses people at the tap, or when a platform partner is measuring you and the number is in a contract. What we refuse: We do not remove header bidding to make a score move. We do not chase a perfect lab score on a page that carries advertising; that number is reachable and it is reached by lying to the test. And we will not take a template where we cannot touch the tag container or the consent script, because that is where the blocking lives and everything else is decoration. Not for you if: Your stack is a managed wrapper you cannot configure and your partner will not open. We can tell you exactly what it is doing to your INP, and then you will be holding a diagnosis you cannot act on, which is a poor use of your budget and our time.