There is a tidal collaboration on this site that openly seeks a second device developer. There is a vision collaboration that asks manufacturers to contribute their rejects. There is a combustion roundtable whose entire premise is that competitors describe their failures to one another.
The instinctive reaction is that nobody will do this. In practice they do, and the reason is that certain questions are unanswerable alone. Array wake interaction cannot be studied honestly with one device geometry. Whether a defect-detection model generalises cannot be established on a single site's data. What a coating does over eight years cannot be discovered by anyone who has not been exposing panels for eight years.
Where the shared question is genuinely unanswerable individually, the calculation changes. You are not giving a competitor an advantage; you are both buying an answer neither could afford, and you are buying it before a consenting authority or a regulator asks you for it.
What makes this work is the terms, and they must be written before anyone contributes anything. Who owns contributed data. What may be published and by whom. Whether participants are named. What happens to a party's data if it withdraws. Whether one participant's product may be ranked against another's — the answer to which, in every functioning collaboration we have seen, is no.
Collaborations that leave these questions until they become relevant do not survive the moment they become relevant. The ones on this board state them on the page, which is why we ask hosts to write them out in full rather than referring vaguely to an agreement to be drafted later.