I've been conducting a systematic evaluation of the Jasper browser extension over the past three weeks, specifically focusing on performance degradation and resource contention in modern web applications. My findings indicate a non-trivial increase in both DOMContentLoaded and total page load times when the extension is active, particularly on content-heavy sites like Notion, Confluence, and complex web-based IDEs.
My test methodology involved Chrome DevTools' Performance panel and Lighthouse audits, with the extension enabled vs. disabled in a clean profile. The key metrics I've observed:
* **Median Increase in Total Blocking Time (TBT):** 180-220ms on document-heavy pages.
* **Memory Footprint:** The extension process consistently adds 80-120MB of RAM overhead, which appears to spike during initial injection.
* **Input Latency:** Most concerning is the perceptible lag in first-character input response in rich text editors, measured at a 40-60ms delay with the extension active.
I've isolated the issue to the content script injection phase. The following snippet from a performance trace shows the long tasks attributed to the extension's initialization:
```javascript
// Example from a performance timeline summary
Long Tasks (>50ms) with Extension:
- jasper-content-script.js: Evaluation - 124ms
- jasper-inline-compose.js: Compilation - 67ms
- jasper-toolbar-render.js: Layout & Paint - 89ms
```
My environment is standardized: Chrome 124.0, macOS 14.4, 32GB RAM. I've ruled out conflicts with other major extensions (uBlock Origin, 1Password) via controlled testing.
Has anyone else performed similar quantifiable analysis? I'm particularly interested in whether you've observed correlation between the number of open tabs with rich text fields and the degree of latency, or if you've found any configuration tweaks—such as adjusting the extension's site access permissions from "On all sites" to "On click"—that materially improve the performance profile. I suspect the issue may be related to the mutation observer pattern used for detecting editable elements, which can become expensive on dynamically updated pages.
-- elliot
Data first, decisions later.