There has been a lot of confusion on the platform, specifically relating to the challenges, on how and when to combine factors. Tl;dr: a non-additive combination of factors of different styles or data sources is highly encouraged. However, we prefer to do additive combinations of factors of different styles ourselves.
Let's say you are building a factor for our insiders challenge. It's okay to submit a factor that is the aggregate of individual factors. But, as the rules state, all individual factors need to be based on insider data. However, this does not mean that insiders is the only data source you are allowed to use for your factor.
So then, could you have a factor that uses multiple data sources — insider and estimates, for example? The answer to that question depends on how you combine them. If the combination is additive (like an average or weighted average of the two), then no, because that's like combining two different factors with assigned weights — it would not be "pure" anymore. In the extreme case, you could just set the weight of the insider contribution to 0.001, fulfill the technical criteria of having an insider-based factor but its contribution is effectively zero. If, however, your combination of these data sources is non-additive with an economic rationale (like an if-else statement, or an interaction) then combination is strongly encouraged. For example, if you think that if insider and estimate sentiments disagree it is predictive, then that is an interaction with economic rationale because you would multiply (rather than add) the positive insider and negative estimate sentiments to measure the divergence between insider and estimate sentiments. This is a type of legitimate alpha that does not come from diversification but from an insight you had — it would be difficult to recreate your factor by just combining an insider and estimates factor together.
In the end, we would like to emphasize that this post is not to suggest you brute-force non-additive ways to combine data sources without economic rationale. How to combine different data sources should always serve your alpha ideas.
If you then have multiple of these insider factors (that are either based on insiders data alone or on non-linear combinations of insiders and other data sources) it is fine to combine these additively, because they are all insider-based (same style) and you'll have one, diversified insider factor.