When printing floating values Zabbix prints value with 15 digit precision, then converts back to float64 and compares with initial value. If those values are not equal the value is printed with 17 digit precision. This appeared to work fine when testing with floating constants. However apparently it's not true for calculated values.

Now printing 0.00003 with 15 digit precision would give 3E-05, which can be converted back to the initial value and is  correct result. However printing 0.00092 - 0.00089 with 15 digit precision gives 3.00000000000001e-05, which throws off our printing algorithm.

One way to improve would be to limit the precision to less digits, however that might result in discrepancy between visual and real data.