Looks like the big difference is due to the addition of a 4MB GIF as the new hero image.
On your before test, you can see in the filmstrip that the LCP is triggered by your hero image, in this case, a 186kb file.
In the second test, when LCP fires much later, we see the hero image is now a 4MB GIF that takes much longer to arrive.
We do track detailed LCP related information and store it in the JSON results for each test, so you can always find out more information there too. For example, here’s what I see for your later test.
This is a kind of interesting example, though. Enough of the GIF is displayed that visually, it looks like the hero image is in place about 8s earlier than LCP fires, but because LCP waits until the an image has fully arrived to to fire, that means the metric gets pushed out quite a bit.