The variables_refreshed_each_compilation acceptance test can fail when the uptime seconds that the test is comparing happen are for strings where the earlier uptime is lexicographically later than the later uptime. For example, this test failed in the puppetserver pipeline here with the following console output:
The assertions are made from here. The values of the variables being compared are strings. In this case though a string value of "9999" would be lexicographically greater than a string value of "10002", so the test fails. The test should be changed to do numeric comparisons of the uptimes in order to avoid the problem.