What is the accuracy and principle of the phpA-B split testing class?

Question

PHP

Good afternoon! There is a need to try the phpA-B
split testing class . I looked, unfortunately I didn’t find reviews and sensible descriptions of the results of working with this class, I’m looking for those who worked with it to answer the question: how accurate is the division obtained with two-variant split testing? If I understand correctly, the breakdown occurs randomly:

...	
  srand((double)microtime() * 1000003);
  $this->current_variation_key = array_rand($this->variations);
  $this->current_variation = $this->variations[$this->current_variation_key]['name'];
...

Tests will take place periodically, but for a short period of time (don't ask why, such is the need / conditions) and therefore the accuracy of splitting in half is important. It is worth using this class in this case, or it is better to store the value of the last issued option (A / B) and, in the absence of cookies from the visitor, change it to the opposite one, thus obtaining the exact order of options (with an error of 1, depending on the end of the period and even / odd last user)?
GoogleAnalytics is also not to be offered, specifically the server version of testing is of interest.

Answer 1

Given the theory of probability - the distribution in the first case may not be exactly 50/50.
But here you need to clearly understand 2 very important things:
1) When split-testing, you need to measure not the number of target actions, but the conversion as a percentage.
That is, for example, case A was shown 2000 times, 20 purchases were made, the conversion was 1%.
case B was shown 1000 times, 15 purchases made, conversion 1.5%.
That is, case B is 50% better than case A.
2) Any split tests make sense on a large sample. You can read more in Google or liraltd.com/raschet-dlinyi-vyiborki-dlya-testirovaniya