Summary

  • Tobias Fiebig and Franziska Lichtblau used a 26 October RIPE 93 plenary to show how trusted expertise can collect undocumented knowledge and exception privileges.
  • Their practical tests are ordinary and uncomfortable: make production changes through the same controls, let someone else run the handoff, and make correction cheap for junior staff.

The fastest person in a network operations room can become its slowest handoff. Everyone knows whom to call; nobody knows where the notes are. The keyring grows. So does the applause. Then the holiday calendar opens and the engineering plan turns into a séance.

In their RIPE 93 opening-plenary talk, Tobias Fiebig and Franziska Lichtblau examined the expert who is right so often that a colleague's challenge begins to sound like a nuisance. Their original slides describe a progression: knowledge stays in one head, documentation gets skipped, and that person becomes the default owner of the next project. Deference is earned at first. Eventually it becomes part of the infrastructure.

The sharpest slide is a production-control thought experiment. One veteran edits a live machine by hand because they know exactly how it works; everyone else uses the formal CI path. The exception has quietly made the veteran the real change policy. The configuration may work tonight, but the team now has two rulebooks and one of them cannot be audited by reading the repository. The speakers' instruction is blunt: expertise buys no exemption from process.

The social part is less easy to put in a dashboard. The deck shows a room of accomplished people either agreeing so readily that nothing new is built, or arguing so well that nothing new is built. It also says rank changes the price of disagreement. A junior engineer takes a career risk when correcting the person who signs off the work; the expert risks little by dismissing them. Silence is therefore a poor measure of confidence.

Fiebig and Lichtblau offer tests that expose the dependency while there is still time to do something about it. Let colleagues own a task and survive a small mistake. Take leave and see whether a documented handoff actually runs. Record a prediction before an intervention, then compare it with the result. Give feedback with a concrete example and its effect. None requires a heroic new platform. Each requires the hero to surrender a small monopoly.

A network can have excellent engineers and still build a brittle chain of authority around one chair. The question after this talk is simple: when the person with every key goes home, which decision still has an owner, a record and a second pair of hands?

Original sources