Searching for "Unable to obtain risk analysis from sippy after retries" indicates that sometimes the Risk Analysis request fails (which of course does not fail any tests, we just don't get RA for the job). It's pretty rare, but since we run a lot of tests, that's still a fair sample size.

Found in 0.04% of runs (0.25% of failures) across 37359 total runs and 5522 jobs 

Interestingly, searching for the error that leads up to this, "error requesting risk analysis from sippy", leads to similar frequency.

Found in 0.04% of runs (0.25% of failures) across 37460 total runs and 5531 jobs 

If failures were completely random and only occasionally repeated enough for retries to all fail, we would expect to see the lead-up a lot more often than the final failure. This suggests that either there's something problematic about a tiny subset of requests, or that perhaps postgres or other dependency is unusually slow for several minutes at a time.