GoogleChrome / GoogleChrome/lighthouse
PageSpeed Insights intermittently includes unrelated bzr.openai.com requests in network dependency tree
- Dominant language
- JavaScript
- Stars
- 30.8k
- Forks
- 9.8k
- Avg merge
- 1d 14h
- Merged PRs (30d)
- 20
Description
### FAQ
- [x] Yes, my issue is not about [variability](https://github.com/GoogleChrome/lighthouse/blob/main/docs/variability.md) or [throttling](https://github.com/GoogleChrome/lighthouse/blob/main/docs/throttling.md).
- [x] Yes, my issue is not about a specific accessibility audit (file with [axe-core](https://github.com/dequelabs/axe-core) instead).
- [x] Yes, my issue is not answered by [other FAQs](https://github.com/GoogleChrome/lighthouse#faq).
### URL
https://ecosurveys.wales/
### What happened?
PageSpeed Insights mobile tests for https://ecosurveys.wales/ intermittently include requests to OpenAI domains which do not appear to originate from the tested website.
In an affected PSI run, the Network Dependency Tree contained requests to bzrcdn.openai.com and:
https://bzr.openai.com/v1/sdk/events?pid=N6fpsoG982MY2Hhy5uZhtc&st=oaiq-web&sv=0.1.41...
That run showed a maximum critical path latency of 2,632 ms and a performance score of 51.
The same site, with no changes between tests, also produces clean PSI runs with no OpenAI requests and a maximum critical path latency of around 428 ms.
I cannot reproduce these OpenAI requests locally. I tested approximately 10 normal Chrome reloads with the Network log preserved and no OpenAI/BZR requests appeared. A Performance API search for openai|oaiq|bzr also returns no resources.
The site does not use a Google Tag Manager container. The only Google tag present is a normal GT- tag.
Local DevTools Lighthouse also does not show these OpenAI requests.
The OpenAI requests are not present in every poor PSI run, so I am not suggesting they explain all performance variability. The issue being reported here is specifically that unrelated OpenAI requests intermittently appear in PSI's Network Dependency Tree.
### What did you expect?
PageSpeed Insights should report only network requests generated by the tested page and its actual dependencies.
Unrelated third-party requests that are not initiated by the page should not appear in the Network Dependency Tree or affect the reported critical path/performance result.
### What have you tried?
I compared multiple PageSpeed Insights mobile runs of the same unchanged page.
Affected runs intermittently contain requests to bzrcdn.openai.com and bzr.openai.com, while clean PSI runs do not.
I also tested locally in Chrome DevTools:
- Approximately 10 normal page reloads with Network "Keep log" enabled
- No requests to openai.com, bzr.openai.com, bzrcdn.openai.com or oaiq resources appeared
- performance.getEntriesByType('resource') filtered for openai|oaiq|bzr returned an empty array
- Local DevTools Lighthouse does not reproduce the OpenAI requests
- The site does not use a Google Tag Manager container
- The only Google tag loaded is a normal GT- tag, not a GTM- container
The page code was unchanged between the clean and affected PSI runs.
### How were you running Lighthouse?
PageSpeed Insights
### Lighthouse Version
13.4.1
### Chrome Version
_No response_
### Node Version
_No response_
### OS
Windows 10 64 bit
### Relevant log output
```sh
```
Contributor guide
Research direction
Start by comparing affected and clean PageSpeed Insights mobile runs for https://ecosurveys.wales/, focusing on the Network Dependency Tree and the reported critical path. Investigate why requests to bzrcdn.openai.com and bzr.openai.com appear intermittently despite not appearing in local Chrome network logs or DevTools Lighthouse. Done means unrelated requests no longer enter the dependency tree or affect the performance result.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- javascript
- Domain
- performance
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Active
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100