I compared number-flow as fixed by five coding agents against my fork v0.2.1 once more. Performance reports changed with the font, the code went through a review with the authors hidden, and a follow-up experiment ended with keeping the mask default.
Every time the value of number-flow changed, style and layout were being recalculated several times over. This post records the performance change observed after 0.2.1 reordered how the library reads and writes the DOM, and the costs that remain.
Sent the Core Web Vitals measured in an app webview to an observability platform shared by many teams. The webview gave no signal on leaving, and tail sampling, delta-to-cumulative conversion and Tempo queries each dropped values in their own way without raising an error. For every setting that could not be changed, re-measured what was lost in a reproduction environment.
Traced why CLS and INP go missing when the page is left, through the transport of OTel JS and the order in which flush runs.
Checked the conditions under which web-vitals 6.2.3 reports each metric, and why INP goes missing when the page is left right after a click.
In the number-flow fork I built to support old browsers, animations in modern browsers were still recalculating styles on every frame. In 0.2.0 I moved the motion built with additive compositing into transform and opacity keyframes, but costs remain in the mask and in interruptions.
I built offline saving into a Marp slide viewer. This post follows the process from the first attempt to run the existing viewer offline as-is to fixing, one at a time, the problems that deploys, networks, and the presenter window exposed.
How much do you lose when 10MiB of functions you never call ships to the browser? I ran 855 measurements to separate the cost of file size from the cost of code shape. Uncalled declarations cost about 21ms per MiB, while initialization code that runs the moment the file is read took 201ms per MiB on the main thread. With real libraries, merely importing a module was enough to evaluate it.
We supported the iOS and Android WebViews of our own app and of a partner app, and gathered the scattered environment branching into adapters. The post covers how we organized insets, the bridge, and CSS, and the problems we hit with the SSR seed and hydration.
I set out to confirm the 500 ms that part 2 left as the cost of going through the worker, but the hard reloads that would form the control group arrive at under one a day. With a control group built from Playwright and a shaping proxy, going through the worker cost 2 ms on a navigation. Most of the extra time on a post click came from the worker saving the post HTML in the background over the same connection, and the gap between lab and field data was a measurement definition difference caused by 103 Early Hints.
A banner built with framer-motion made the entire home screen stutter for the 0.6 seconds it took to open. My guess at the cause from reading the code was overturned twice by measurement, and in the end a single easing function removed the reflow.
This post covers the service worker material that did not fit into the cache chapter of Frontend Performance Optimization Deep Dive (published in Korean). It works through why caches go stale, why the old version shows up after a deploy, what to store under which strategy, and whether you should use one at all, and measures a 104KB opaque response being counted as 6.6MB of storage.
number-flow turns animations on starting from Chrome 125 and Safari 17.2. While building a fork that lowers this floor to Chrome 66 and WebKit 16.4, I reversed two decisions, and after investigating a Safari bug I gave up on automatic downgrading.
What happens when React wraps the View Transition API
How to efficiently observe hundreds of elements while preventing memory leaks
In July 2025, the browser GPU API changed generations for the first time in 14 years
If you find yourself thinking "maybe I should try this too..." — reach out right now!!
