The answer key was never written down

Nearly all engineering knowledge is public

Nearly all of it. The Eurocodes are published, along with every national annex that adapts them country by country. Trafikverket's technical requirements are published. AMA, the Swedish reference texts for construction work, can be bought by anyone. And because public procurement runs on public records, the deliverables follow: the investigations, calculations, technical descriptions and plans that infrastructure projects produce end up in registries anyone can request. In Sweden, offentlighetsprincipen makes the archive of delivered engineering work one of the most open in the world.

This openness is easy to mistake for completeness. If the method is in the code and thousands of worked examples sit in public archives, what is left to know? Every practising engineer can answer that. The distance between reading the code and daring to sign the deliverable is the profession itself. The rules are the textbook. And nobody passes the exam with only the textbook.

The archive stored the answer, not the correction

The archive stores final documents. A signed report shows the conclusion: this foundation method, this cross-section, this stormwater solution. What it does not show is the path there. Somewhere between the first draft and the signature, a senior engineer intervened. An assumption got tightened, a load case added, a value pulled back toward caution. One deviation from the standard was flagged as serious; another was accepted with a note. The final PDF shows none of this. It stores the answer and discards the correction.

The correction is the valuable part. Two deviations can look identical on paper, and an experienced engineer knows that one is routine while the other will surface as a problem two winters into operation. No clause in the Eurocodes says which is which in this soil, on this road section, for this client. That call is what senior review exists for. It was never written down anywhere: it passed from person to person, project by project, the way craft knowledge always has.

There is a second gap. A delivered report proves what was accepted, not what was correct. Reviews miss things, and some archived reports contain exactly the errors the review was supposed to catch. Read every public document perfectly and you arrive at accepted practice, not beyond it. Anyone can train on the rules. Only practice grades the answers. The answer key was never written down.

It matters now because the answer key is retiring

Because the people holding the answer key are leaving. Cedefop, the EU's agency for vocational training, projects that around 4.1 million people will be needed in European construction between 2022 and 2035 to replace workers leaving the occupation, mainly through retirement. That is a projection, not a measurement, and should be read as one. But the direction is hard to argue with: in the same forecast, replacement demand exceeds new job creation by a factor of nearly fifty. The industry's staffing question for the next decade is not growth. It is succession.

The profession has exactly one mechanism for transferring judgment: years of working beside someone who has it. It works, but it is slow, it does not scale, and it leaks. When a forty-year career ends, the judgment ends with it, minus whatever fraction rubbed off on younger colleagues along the way. And it must be rebuilt in every new engineer from scratch, one project at a time, while the volume of documents each project produces keeps growing and the number of people who can judge them does not.

Now the judgment can be captured as it is exercised

For the first time, the judgment can be captured at the moment it is exercised. When a machine drafts a piece of engineering work and an accountable engineer reviews it, every decision that engineer makes is a verdict on a real problem: accepted, corrected, rejected. Not an interview about how they work. Not an annotation job on invented examples. Real deliverables, real site conditions, graded by the person whose signature carries the liability.

A verdict is a different kind of material from data. The reports were always public, and the AI labs can train on all of them; that takes a model to accepted practice, the same ceiling the archive sets for everyone. The verdict is what the archive never held: whether an answer was right, in this context, in the judgment of someone accountable for it. Collect enough verdicts and the unwritten answer key begins to exist, not as a book but as a growing record of what accountable engineers accepted, corrected and rejected on real problems.

That changes what retirement means. Judgment exercised this way stops evaporating at the leaving party. Every verdict a senior engineer passes keeps working after they stop, and every younger engineer starts on top of the accumulated corrections instead of rediscovering them one mistake at a time. The profession's most perishable asset becomes its most durable one.

This is the principle Yesper is built on. Yesper is the AI civil engineer for construction and infrastructure, and it learns from the verdict, not from the data: what makes it better is the accountable engineer's judgment on real work, never the customer's documents.