A chat or review widget can look small while adding scripts, frames and main-thread work. Commission an isolated before/after comparison with a stated business benefit and a fallback. File size alone cannot tell a buyer whether interaction became slower.
Primary evidence and claim boundaries
Google’s web.dev guidance describes third-party JavaScript costs including network requests and main-thread parsing/execution that can delay interaction. Primary reference · Read 17 October 2026.
Filled acceptance example
Synthetic teaching scenario, not a client case, provider integration test or business result. Scroll the example table on narrow screens.
| Synthetic controlled observation | Baseline → widget | Review meaning |
|---|---|---|
| Transferred script bytes | 80 KB →260 KB | +180 KB; compression/cache context required |
| Main-thread task duration in chosen trace | 120 ms →310 ms | +190 ms in this sample; not an INP score |
| Chosen project budget | Additional script≤150 KB | Example exceeds budget by 30 KB |
| Widget fails to load | Static contact link retained | Business path remains available |
Practical handover procedure
- Agree the task the widget solves, the owner and a synthetic or authorised isolated test page. Keep content, browser, network, cache state and device conditions comparable.
- Measure several runs with the widget enabled and disabled. Record distributions and trace evidence for loading and interaction; do not substitute a single Lighthouse score for all user experience.
- Compare against the agreed budget and functional benefit. Evaluate supported delayed-loading or a click-to-load facade on the isolated page, including its accessibility and failure behaviour.
Interpret the outcome
The filled arithmetic finds a 30 KB budget overrun, but it does not claim a 190 ms real-world regression or any effect on enquiries. The final procurement decision needs repeated comparable observations and functional evidence. A facade is a candidate design, not automatic permission to change a production page.
Acceptance decision
Require a reproducible setup, repeated measurements, business-function checks and a working failure path. If there is no reliable baseline, state performance impact unknown.
Limits and unverified work
All measurements and budgets here are synthetic. No production performance repair, homepage change, new tracker or third-party script was executed.
Download the filled bilingual acceptance record — Contains these inputs, steps, expectations and limits. This is a teaching record, not an executable integration.