INP replaced FID in 2024 — what changed and why your old Lighthouse score lies

On 9 September 2024, Google retired FID (First Input Delay) and replaced it with INP (Interaction to Next Paint) as a Core Web Vital. The difference is fundamental:

• FID measured only the delay from a user's first interaction (click, tap) to the browser starting to process it. After that, nothing.
• INP measures the full interaction window — from user input through event processing, JS execution, layout, paint — for the slowest interaction across the entire session.

Why this matters

INP catches what FID missed: heavy on-click handlers, expensive re-renders, layout thrash from CSS-in-JS frameworks, and the dreaded "tap-to-respond" lag on mobile-heavy SPAs. A site that scored "Good" on FID (under 100ms) frequently shows INP in the 300-800ms range — visibly slow.

The thresholds

• Good: INP ≤ 200ms (p75)
• Needs Improvement: 200ms < INP ≤ 500ms
• Poor: INP > 500ms

How to fix INP regressions

1. Break up long tasks: anything > 50ms blocks the main thread. Use scheduler.yield() or setTimeout(0) to chunk.
2. Move heavy work to Web Workers: parsing, encryption, image processing — anywhere CPU is the bottleneck.
3. Audit React re-renders: memo, useMemo, useCallback, and React 19's compiler help. Big lists need virtualization (react-window).
4. Defer non-essential JS: third-party scripts (analytics, ads, chat widgets) should load async or defer — never in critical interaction paths.
5. Trim CSS-in-JS runtime: prefer compile-time CSS (CSS Modules, Tailwind, vanilla-extract) over styled-components for SSR pages.

Measure both lab and field

Lab (Lighthouse) gives you a snapshot. Field (Chrome User Experience Report, CrUX) shows the p75 across real users — what Google ranks on. AuditOPE pulls both via PageSpeed Insights API + lab Lighthouse 12.6.1, surfacing INP as a separate metric with the score, threshold, and field-vs-lab delta.

If your performance reports are still showing FID, your tooling is 18+ months out of date.