I wouldn't characterize Trellis and NSS quite in this way. One of the basic design goals of Trellis is to implement *existing* web specifications. There is always a danger in implementing something that hasn't been specified or for which the specification is still a moving target, and given that the Solid specification hasn't yet reached First Public Working Draft, it is still way too early to talk about Trellis implementing the Solid specification. So for now, the goal of Trellis is to implement the LDP specification. As such, it isn't especially difficult to extend Trellis or any LDP server so that it also can work within the Solid ecosystem. In fact, in many ways, Trellis already interoperates quite well: the differences tend to be in the areas of authentication rather than storage/retrieval. It is also worth mentioning that, while there is a very large area of overlap between Solid and LDP, LDP is not strictly required by Solid. That is, not all Solid Servers will be LDP servers, and not all LDP servers will be Solid servers. At this stage, I am very interested in ensuring that Trellis doesn't make it hard (or impossible) to work with the Solid ecosystem, but that's different than having an explicit goal of implementing a specification.
On the other side of this story is NSS, which existed long before the current effort to formalize Solid as a specification. That specification effort began in earnest just over a year ago. NSS began
6 years ago. The specification effort clearly builds on the implementation experience of NSS, but it is not bound by it. For example, the way NSS implemented authentication is not being carried forward. The Solid specification may even leave NSS behind at some point, depending on the extent to which the maintainers of that software choose to align with the Solid specification. Some of the people who have been very involved with NSS have chosen to start a brand-new
Node.js-based implementation of Solid rather than put more energy into NSS. Having spent a lot of time discussing the two codebases with the authors of the new community-server, I would put considerably more trust in that new code base than in NSS, even though the community-server is far from complete. And in terms of specification conformance, the community-server is what you should look at as a public reference implementation of Solid.