Of the three questions, two will be answered by the market without anyone having to fight for it.
Whether a system's behavior is written down where a machine can check it — that one commodifies. It's a hard engineering problem, but a tractable one, and tractable problems get solved and then get cheap. Assume that within a few years everyone who cares to will be able to measure it, the way everyone can measure test coverage today. It will be table stakes, not an edge.
Whether behavior has been lost — the unaccounted-for, the tripwires — that one yields to honesty. You can't measure it directly; what's truly lost leaves no record to read. But it casts a shadow. It shows up in the incidents that land exactly where nothing was watching, and anyone willing to keep looking can bound it, corner it, make it smaller over time. It takes discipline, not invention.
The third question does not behave like either of them. Can a human still predict what this system will do, find where a change has to be made, and defend why it's built the way it is? Whether anyone still holds the living understanding of the thing — that one resists, and it resists for three reasons that stack on top of each other rather than sitting side by side.
The first reason is the one this whole series has been circling. You cannot recover this by looking at who wrote the code. That was the natural place to look — for fifty years, the person who authored something understood it, so authorship was a usable fingerprint of understanding. That fingerprint is exactly what the last few years erased. The author of a system may understand it completely, partly, or not at all, and the record looks identical in every case. The one inference everyone reaches for first is the one that no longer holds, which means the answer has to come from somewhere harder to reach than a commit log.
The second reason is that it moves no needle on the way down. The other failures in this series at least announce themselves eventually — an incident, a missed behavior, a number that finally turns red. The erosion of understanding makes no sound. Nothing in the daily readout dips as the last person who understood a subsystem gradually stops thinking about it, or leaves, or simply forgets. There is no slow slide you could catch if you were watching. It is fine, and fine, and fine, and then it is gone — and you learn it's gone at the moment you reach for it, which is the one moment you cannot afford for it to be missing.
Which is the third reason, and the cruelest. This is precisely the capacity you need at the worst possible time — when something has broken, or has to change under pressure, or has to be explained to someone who is right to demand it. Understanding is the thing you draw on in exactly the situations where there's no time to rebuild it. A reserve that is always full except on the day you need it is not a reserve. It's a surprise.
Put the three together and you get a situation most organizations are walking into without a name for it: a system that is fully verified — every test passing, every behavior anyone captured still holding — and understood by no living person. Not a broken system. A working one. It runs, it's green, it ships. And there is no one left who could tell you why it does what it does, or what would happen if you changed it, or which of its behaviors are load-bearing and which are accidents that have simply never been disturbed. That is not an exotic failure. It is the ordinary, default destination of producing faster than you understand, left unmeasured, for long enough. It is where the road from the first essay leads if no one is reading the gauge.
You will not learn this from who wrote the code, from whether the tests pass, or from how fast the team is shipping. Those are the three places everyone is looking.
So the question — can a human still account for this system — is at once the one that matters most and the one almost nothing in the industry is built to answer. It is the axis that actually fails, it fails silently, and it fails at the worst time. It has a name: legibility. Naming it is most of the work, because once it's named, the absence of any instrument pointed at it becomes impossible to un-see.
But naming a thing is not the same as reading it, and a claim this specific — that the one axis that matters can be measured, when the obvious way to measure it is the very thing that broke — cannot be left as an assertion. It has to be built and shown working, or it's one more exhortation in a field already full of them.
So it was built.