Too small a sample to tell. Sorry, because when is enough, enough? 30x100 is not enough ( sorry Charles), or 40 x 100.
Intriguing. Suggest you prepare several more equivalent sample tests. Say for example 7 sets of 4,000 and compare.
10,000 spins might be a better single sample module, and then compare say at least 3 samples. I have one colleague who generated several million RNG spins to test his method which has a family of bets showing about +2 to +5% edge.***Such a small degree of tuning, but it needed vast samples 'to be sure'. He now plays the methods live ( requires comparative cycle analysis from 100 to 300 spins - takes a huge amount of patience - not for me). He plays now with larger value units live and has been doing this ( despite a few personal behaviour blips/ meltdowns- these can be trained out) for the past 12 months. We meet several times a week at the casino.
Also compare live vs RNG. Why would one result differ so much? One suggestion is that your bet choice is very 'transparent'/ legible ( easily read) to the RNG program so as it is designed to beat you ( this is a fact) you cannot camouflage your bet. My own bet methodology is harder to read because it involves multiple spreads of 9 number targets. In theory if I had say six or seven sets going simultaneously the poor program would become confused because all numbers might be covered at any one time ( but of course and even though flat staking ) were at different stages of a six or seven step bet method ( earning from +27 to -27).
However I am keen to see the sort of bet you placed and why the RNG seemed to be so comparatively negative, or perhaps over a set of several samples these results might be equivalent between live and RNG. However I suspect the simple flat bet RNG result is doomed.
I was surprised nevertheless with a very small RNG sample test I did recently at how the bet dealt with RNG ( just like any other source) but it was a ridiculously small sample of only 98 spins ( although spread over 4 sets ( so 4x 98). Still way too small to form any valid opinion.
What actually were you seeking to demonstrate? It has raised several issues so thanks Hans.
*** The flaw with the approach from my friend S. is that his generated samples in RNG were not generated in response to any bet. They were neutral. In live RNG play the program will generate results in response to what you lay - takes only micro seconds to process. S used to play online in a shadowy style where there would be other players at the table, and so he could 'hide' behind their smoke screen. This will not work in the live RNF in the casino as he may be the only player active, and he has been caught out there in this manner, so I think now prefers the Rapid Roulette live table. I will check to see his latest comments. His bet is vulnerable because it may only expose say 3 to 5 targets occasionally.