Technical Sharing/Website Performance
Over the past decade, website performance budgets have largely revolved around loading: resource volume, number of requests, image and script compression, and first-screen drawing time. This approach has a clear indicator correspondence-LCP (Maximum Content Drawing), as well as mature compression and clipping methods.
Interactive response is not in the budget table for a long time. Whether a page can be ordered, entered or not followed, expanded and submitted is usually regarded as an experience problem of "adjusting after development", rather than a constraint of allocating quota before going online.
The number of page interactions has also increased over the years: single-page applications, dynamic filtering, instant input feedback, and step-by-step form submission all put the response pressure on the interaction side. Loading is slow, the user can usually wait;If the interaction does not follow hands, the user will directly judge that this page is not useful.
We call this amount the interaction budget: the amount of response time reserved for high-frequency interactions on the page, alongside the load performance budget. Its measurement indicator is INP(Interaction to Next Paint, interaction to next drawing), which covers all interactions such as click, input, expansion, submission, etc., not just one click.
The boundary should also be clear: the interactive budget is only interactive response and does not include resource volume and load speed.
Load Budget and Interactive Budget Side by Side
The first group is the August 2026 crUX data set, with 18,294,881 sources: LCP good 68.1%,CLS 81.5%,INP 85.3%, and all three items over 55.6%. INP has declined month-on-month since March 2026, and the Chrome team said in a release note that it has not yet determined why.
The second group is an annual comparison: from August 2025 to August 2026, LCP rose from 66.7 per cent to 68.1 per cent, CLS from 80.1 per cent to 81.5 per cent, INP from 85.7 per cent to 85.3 per cent, and all three items rose from 53.5 per cent to 55.6 per cent.The loading class indicator is improving and the interaction class indicator is falling back.
The third group is equipment grouping. In May 2026, the data set and 13.75M sources were all over 49.1 on the mobile side and 58.0 on the desktop side, a difference of 9 percentage points. There are more interactive scenes on the mobile side, with input, sliding and unfolding on this side, and the response pressure is also greater.
According to Google's indicator threshold, INP is good within 200 milliseconds, needs improvement 200 to 500 milliseconds, and is poor over 500 milliseconds. This threshold is the scale on which the interaction budget can be compared.
The combination of the three sets of numbers means that the optimization on the loading side is close to the regular action, while the performance on the interaction side is falling back, and it is precisely the part that the user feels most directly.The differentiation of indicators indicates that the problem is not in overall performance, but in interaction, which is not managed separately.
The interactive budget addresses the speed of response and does not address whether the function is easy to use or whether the information is clear; nor does it replace usability and accessibility checks.
What is easy to confuse is to attribute interaction issues to visual or development scheduling. Interactive response and what the interface looks like are two things: the former depends on script execution and the main thread occupation, and the latter is style and layout.
before going online, you can ask yourself four sentences: are the high-frequency actions listed? is there a response time limit for each action? have you tested it on a real device? when was the last retest?
The STOYARD writes the interaction response and loading performance together into the experience constraints in the Define phase, the Build phase verifies the response of high-frequency actions on real devices, and the Evolve phase retests the core metrics on a quarterly basis.