I started looking into merging RuntimeStatsTotalMetric and RuntimeStatsMetric. 
Currently, RuntimeStatsMetric considers all the samples till time to interact (TTI) or  DomContentLoaded time (https://cs.chromium.org/chromium/src/third_party/catapult/tracing/tracing/metrics/v8/runtime_stats_metric.html?q=RuntimeStatsMetric&sq=package:chromium&l=114). The idea was to measure the startup performance of the benchmark. 

We were planning to use the "load expectation" bucket as a measure of the startup performance. However, the "load expectation" measures only time till first contentful paint (https://cs.chromium.org/chromium/src/third_party/catapult/tracing/tracing/importer/find_load_expectations.html?sq=package:chromium). In my understanding First contentful paint happens quite early in the navigation and may not be very useful to measure the startup performance of V8.

The other buckets in the UE category are: Idle, Input and Startup (this is the startup of the thread). None of them seem to be good to measure the startup performance.

Any suggestions on anything else I could use? 

We could add other metrics like TTI_Compile, TTI_Total that include all the samples till the startup, but I am afraid it might be a bit confusing. Since we would then have Total_Compile = (Load_Compile + Idle_Compile + Startup_Compile + Input_Compile) and then TTI_Compile which is not a sub category.

I changed load expectations to use TTI, so now we could use the load bucket to measure startup time. Though on some of the pages (for ex: cnn on cold run) we don't generate the initial navigation start event and hence we don't produce a LoadExpectation on these pages. Hence the load UE bucket will not have any samples.