- From: Nicolás Peña <npm@google.com>
- Date: Tue, 15 Oct 2019 18:04:38 -0400
- To: public-web-perf <public-web-perf@w3.org>
- Message-ID: <CAAATDin63mfwMG_i_Ex_E65uEPAKrPeGULyTkp9wnVcBecOQPg@mail.gmail.com>
Hi all, At TPAC, some concerns were raised about the usefulness and the heuristics of the LargestContentfulPaint (LCP) metric and publically available data was requested to better evaluate the metric. Here is a summary of the main points and what we intend to work on. Let us know what you think! 1. LCP excludes background images of the <body>. There were concerns about this heuristic being arbitrary. To address this, we will present sample filmstrips where LCP is different due to removal of background images of the <body>. 2. There was a request to look at how LCP correlates with ‘user happiness’ and business metrics. Chrome does not have a plan to obtain 'user happiness' and none of our metrics intend to measure this. We could try to correlate LCP with engagement metrics, but the correlation might not be there: if all you want is to get some information quickly, a fast LCP might make your interaction with the page faster, but that's not necessarily bad. We'll be looking into correlation with business metrics. 3. Chrome has not shown that LCP is a reasonable metric by correlating with known ‘good’ metrics, such as SpeedIndex. We will do a correlation study of LCP with SpeedIndex, FCP, onload, and maybe others. We should expect correlation primarily with SpeedIndex, but it would be interesting to look at correlation with the other metrics. Perhaps LCP can be viewed as a RUM alternative for SpeedIndex. 4. Chrome has not shown that LCP is a necessary metric (i.e. not redundant, given the metrics developers could compute nowadays). The correlation study can help shine some light into this. If there is not a good correlation between LCP and other metrics that can be measured in the wild, then that’s a good signal of independence. In addition, we can show examples where LCP differs from the page load metrics that are easy to compute in the wild today (FCP, onload, any others we should include?). These examples would highlight LCP as an ‘independent’ metric: one that cannot be easily estimated by using the other load metrics available today.
Received on Tuesday, 15 October 2019 22:05:03 UTC