Main idea

A tree always works well on the objects it was drawn for. You know where to file a source, a note, a decision, a deliverable, because a place has been planned for each. Checking that every known file has a slot therefore measures nothing: it amounts to checking that a plan covers what it was used to establish.

The trial comes with the object nobody anticipated. A translation guide didn't exist when a workspace's frame was defined — nobody had anticipated that a book would be translated while it kept evolving, nor that those translations would call for a set of rules stable enough to carry across several languages and several sessions. The question then becomes: do the distinctions already laid down allow a decision, or does a new one have to be invented for the occasion?

An architecture is robust when it correctly files an object it never anticipated. That is the only measure that carries information, because it is the only one the designer can't set in advance.

Why it matters

This shifts what you look for when assessing a filing plan. Completeness — everything has its place — is a comfort indicator; the capacity to absorb the unexpected is a design indicator. The two can diverge entirely: an exhaustive plan built on a list of known objects collapses at the first object off the list.

It also suggests a way of working: deliberately submit a nascent architecture to a case it wasn't used to build, rather than validating it on its original corpus.

Nuances and limits

An architecture can pass the test on one object and fail on another; the result doesn't generalise, it accumulates.

And the trial doesn't say whether the answer is good. It says an answer could be drawn from the existing distinctions — which is compatible with a debatable filing, defended by a criterion that stands up.

Open questions

• Can you deliberately manufacture test objects for an architecture, or do they lose all value as soon as they are chosen by whoever drew the plan?