You do need to define things a bit more. Daily data or tick? Stock, futures, etc. Just US data or more.
How are you going to allow any languages to run on your server? That seems pretty dangerous.
How are you going to allow any languages to run on your server? That seems pretty dangerous.
As for security, we plan on running each user's code in its own sandbox, with restricted permissions.
I'd also suggest getting fundamental data for the equities, so your customers have more than just hi/lo/open/close/volume for their strategies. You'll find that making sure none of this fundamental data is snooped is hard. (i.e. making sure the data isn't revised back in time, so it is as it would appear on a given date). That messiness is your good value add if you can get it right. Also, please encourage out of sample testing using some form of cross validation.
To your point about back testing, would you want guidance as to what the in-sample and out-sample time ranges should be? Tools to help you make sure you're not overfitting? Or something else?
I guess I'd want an easy way to get data in chunks for testing. You can either do it as a rolling window or regular cross validation. (Rolling would be train on N years of data, test on the next year. Then drop the last year and add another year and repeat. CV would be just divide in 10 chunks, train on all combination of 9 and test on the 10th.) It would be good to support both.