Github user kaspersorensen commented on the issue:
https://github.com/apache/metamodel/pull/128
I updated your gist. I think there is still a good case for using decimal
avg's ... Here's my output, indicating that your test is more often wrong than
right when applying more decimal precision and real rounding instead of
int-division (which I believe always rounds down):
```
false alarms: 2693
errors: 149, total tests = 1000000000, percentage: 0.000015%
```---
