The failure lives in the join, and no single vendor owns the join.
I had four hours in a terminal I did not know, which is the ideal condition for looking at wayfinding, because you are simultaneously the intended user and an observer with nothing else to do. I filled about twenty pages. Almost none of what I wrote down was about type.
The terminal's signage is genuinely good, and its quality is almost entirely a matter of sequence. Decisions are presented one at a time, at the point where the decision is made, and never earlier. There is no sign anywhere that lists more than three destinations. Crucially, the sign that resolves a decision is visible from the point at which you commit to it — you can see that you were right about four seconds after you turn, which is the interval at which doubt would otherwise start.
Where it fails, it fails at handovers. The signage system stops at the airline gate area and a different system, procured separately, takes over. The typeface changes, which is trivial, but the logic changes too: the second system numbers gates in a sequence that runs the opposite direction. I watched four people walk the wrong way in ten minutes at exactly that seam. No individual sign in either system is bad. The failure lives in the join, and no single vendor owns the join.
The other failure is temporal. Every screen shows departure time and gate; almost none shows walking time. The single most useful piece of information for a person with a connection is how long it takes to get there, and it is the one number that the system, which knows the distance perfectly well, does not compute. This is not an oversight of design so much as of scope: the distance lives with facilities, the departure time lives with operations, and nobody was asked to join them.
I flew home thinking about the handovers rather than the signs, which is roughly what happens every time I look at any system carefully. The typography is nearly always fine. It is the seams, the sequences and the things nobody owns that break.