When investigating an issue, we discovered that while FCP is properly ignored when the document is fully opaque (e.g. in cases where an A/B anti-flicker snippet is deployed), it does not necessarily fire when the document's opacity changes, unless there are paint involved with that opacity change. One particular case where paints are elided is when a will-change: opacity style is applied to the document.
If your site is using an A/B testing framework that employs full-document opacity snippets, then this issue may have affected your site's FCP metrics.
The issue results in an FCP entry that indicated the time of the next contentful paint after the opacity change, which could be significantly after the actual FCP. Potentially, there could also be cases where FCP wasn't reported at all.
This seems to have been a long standing issue, most likely introduced when we started ignoring paints with opacity 0 in Chrome 86.
