While I don't have a definite answer, I anticipate sometime in 2023. I will be conducting small scale testing with select individuals, and the release date will depend on user feedback through those trials and the number of revisions necessary. Signup for email updates to have a chance at early access!
Users will certainly be provided the opportunity to explore the site for free to start. Each additional user, however, results in additional server and compute costs, which I intend to recoup through some type of monitization. I expect that any cost will be more than made up for by the incremental investment returns resulting from the site's content.
Data
All of the exact sources are proprietary, and the list is constantly growing. Some of our sources include the SEC, the Bureau of Labor Statistics (BLS), Nasdaq, Yahoo Finance, Twitter, Reddit, Individual Company Websites, TD Ameritrade, and others.
We have data points dating all the way back to the early 1990's, but not all of our data is used to train our models. Each model is trained using many different ranges of data, and the results are ranked to determine the best training window for each model.
Models
Many different algorithms are used throughout the modeling process. For each model we test different forms (Decision Trees, Recurrent Neural Networks, Convolutional Neural Networks, Gradient Boosted Machines, ... etc.) and select the form that works best for each individual model. For exmple, a recurrent neural networkl performs much better when predicting short term price movements based on time-series price data, whereas a decision tree performs much better when predicting sector and industry trends.
At time of training, standard model evaluation techniques are used, including analysis of AUC, MSE, Confusion Matrix, ... etc. The ultimate evaluation criteria, however, is based on a historical backtest of each model. To perform the backtest, we train each model through a series of different dates and then evaluate the model's predictions against actuals on an out-of-time holdout set. The ultimate evaluation criteria we use compares the actual investment returns realized based on investing the model's most confident predictions vs. returns realized based on purchasing the entire market.