I haven't explored using Facets. Mostly because I haven't seen them used consistently and, in practice, we haven't had much of a need for it (in addition to separating out test code, etc., it might also help in cases where a single Git repository is used to create multiple JARs, but I don't think that this is actual use case for facets). Theoretically, you could use Facets to do your own evaluation (e.g., use the GLOB patterns specified to grab the license data for only those facets that you care about).
NOASSERTION means that the discovered license cannot be mapped to an SPDX _expression_. This identifier is unintuitive as "no assertion" really sounds like "unspecified". From our point of view, it's a blocker that requires further investigation.
The string NOASSERTION. This indicates that license-like data is found, but that ClearlyDefined cannot identify an SPDX-identified license.
ClearlyDefined uses various tools, including
scancode-toolkit to determine licenses (IPZilla uses this tool for third-party content scans). In my experience, the tools are generally good but do make mistakes (this is one of the reasons why we've been pushing the use of
SPDX-Identifier tags in file headers). I'm looking at
one record today that has one file that was incorrectly identified as NOASSERTION (which I will submit a curation for).
So one way to improve the results would be to provide the scancode project with patches. Another is to just provide your own curation. You can make changes directly in the ClearlyDefined data yourself and push them for review by a curator.
There are some cases where we (the Eclipse IP Team) don't agree with what is in ClearlyDefined, have curations that are currently too cumbersome to push, or the data is completely missing from ClearlyDefined. This is why the license tool first checks with IPZilla to see if we're trying to override. For Eclipse projects, IPZilla is still the primary source of license data.
My intention is that eventually we will start reducing our dataset by moving it to ClearlyDefined. One of our challenges here is that ClearlyDefined makes it super easy to push one curation. Pushing multiple curations (e.g., multiple versions of the same content) is possible, but is a very manual process (which involves YAML for some insane reason).
FWIW, the curation that I provided for maven/mavencentral/org.eclipse.jdt/org.eclipse.jdt.core/3.22.0 that corrects the source repository and discovered licenses was accepted. So that's one...