Data quality and accuracy of test results

Hi, I'm starting to test the data quality and I discover a strange beahvour so I decide to did the following experiment:
I train and test workshop6 from 2012 to 2016 with only one asset eurusd.
then I erase EURUSD_2012.t6 - EURUSD_2016.t6 and I download it from fxcm (api)
after retrain i can read a complitly different results:
orginal data

Number of trades 142 (89/year, 2/week, 1/day)
Percent winning 36.6%
Max win/loss 516$ / -337$
Avg trade profit 4.54$ 4.0p (+116.9p / -61.3p)

fxcm data:
Number of trades 65 (41/year, 1/week, 1/day)
Percent winning 49.2%
Max win/loss 1339$ / -925$
Avg trade profit 10.37$ 3.5p (+140.1p / -129.0p)

I can't understand; why this workshop is so data depending?

Can someone help me to understand?

thks a lot, alb

Workshop 6 reinvests profit. Tiny differences thus have large impact on the end result.

Hi jcl, thanks for your answer, for your time and for your great work.
Sorry I forgot to specify that I disable the reinvest ability. I only use optimalF and a fraction of capital like yours examples so I enabled
Margin = 0.5 * OptimalF * Capital;

Can I ask you the source of the data in the setup.exe package and if you act some kind of denoise, correction or filter in the data? I start to coding my own scripts but I need to trust in the results of test, for avoid wrong belief.
many thanks for your help, Alb

As far as i know, i think zorro provided data is from Oanda.

Data from different providers can have significant result differences, especially in forex and CFD, not to mention commisions, slippage and so on.

Surely jcl can give you more details.

I found no records of the data source, but if I remember right, the Forex data before 2017 was from Oanda and the CFD data from FXCM. If you want to trade with a particular broker, it is recommended to test the strategy with data from that broker if available. For the workshops it won't matter much since they are for learning only.

alb,

would you be able to post 2 charts - same time window - one for each data set, both showing at least 1 trade each?

thanks alot for yours answers,
@luisd no I can't do this experiment because I'd aligned all t6 files in all zorro installations and I can confirm the same results in this conditions as expected.

thks

I have deepened my studies and I realized that some indicators are more susceptible to the small variations caused by the otc market. For example, the broker's management of weekend candles affects the atr in a significant way. However, even if it's a little rough, I'm using this method to test the robustness of my scripts and I have to say that by making small changes the scripts become less susceptible to changes (for example by applying indicator smoothing)
I also noticed a big difference between the currency data and the index data provided by differents brokers.
Finally it is impressive note how weakly the z12 script is susceptible to data variations