Concept Space hands you the grid but not the axes. You name the dimensions yourself, which is the part that takes a little thought, because a grid built on weak axes will look busy and tell you nothing. Three properties separate an axis that does work from one that just decorates the screen.
Axes should be roughly independent
If two axes move together, the grid collapses onto its own diagonal and half the cells can never be filled. Suppose you set one axis to price and another to perceived premiumness. These are nearly the same dimension wearing two labels, so almost everything lands on the line from cheap-and-cheerful to dear-and-fancy, and the off-diagonal cells, expensive but positioned as humble, cheap but positioned as luxury, sit permanently empty for reasons that have nothing to do with your portfolio. You learn nothing from gaps that are gaps by construction.
Good pairs are closer to orthogonal. Audience reach against tone is a strong pair, because a niche thing can be playful or serious and so can a mass thing; all four corners are genuinely reachable. When you are choosing a second axis, ask whether every value of it is compatible with every value of the first. If the answer is no, you have picked two views of one dimension.
Axes should have a few discrete values
A quality dimension in Gardenfors' sense is continuous, but a grid you can read needs that dimension cut into a handful of bands. Three to five values per axis is the workable range. Two values gives you a two by two, which is sometimes the right tool but rarely rich enough to show clustering. Eight values gives you a grid too sparse to read, where almost every cell is empty simply because you do not have eight times eight projects.
The cuts should be ones you can apply confidently. Audience as solo, small team, whole org is easy to place a project into. Audience as a raw number of users is not, because you will dither over the boundaries and the placements will not be stable. Name the bands in words, make them mutually exclusive, and make sure you can drop any real project into exactly one of them without a long argument with yourself.
Axes should matter to a real decision
This is the test that retires the most candidate axes. An axis earns its place only if the position of a project along it would change what you do about that project. Delivery mode, web, native app, CLI, physical object, matters because it changes who can reach the thing and what it costs to build. Founding year usually does not matter, because knowing a project is old or new rarely changes your next move, however easy it is to record.
A quiet sign that an axis fails this test is that you find it interesting but cannot say what you would do differently if a project were at the other end of it. Interesting is not the bar. The bar is decision-relevant, because the whole reason to find the gaps is to decide whether any of them is worth filling, and a gap along an axis that drives no decision is not worth your attention.
A practical starting set
If you want somewhere to begin, audience reach, tone, and delivery mode make a solid first three: independent enough that all the corners are live, each naturally cut into three or four named bands, and each tied to a real decision about who a thing is for, how it should feel, and how it gets to them. Map your existing work across those three pairs first, see where you have unknowingly clustered, and only then start adding the axes that are specific to your own situation. Better to start with three axes that pass all three tests than ten that mostly do not.