Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Main idea
A workspace can bring together the standards of the field, the product documentation, customer interview transcripts, in-app feedback and properly tagged support tickets. On paper, nothing is missing any more: every family of material has been collected, sorted, made readable.
What is missing, though, is what the product manager carries in their head. Not a hidden theory — an accumulation of small things: a nuance grasped after three exchanges with the same customer, a fear that cannot be heard in a transcript because it was never said, a political tension between two departments that nobody will ever write down, an interpretation still too fragile to defend in a meeting.
Documentary completeness and completeness of understanding therefore do not overlap. A context is judged on the second, and nothing in its inventory of sources lets you observe it.
This lack, long noticed without being measurable, has an instrument: hand an evolution to an agent and look at the gap between what you wanted and what you get back. That gap is exactly the part of the model you had not written down. The list of holes then stops being guesswork: it is read in the rework.
Layer added by "I wrote a product's ontology. Three times, I thought I was done." (2026-08-11).
Why it matters
This shifts the diagnosis when an AI plugged into a well-kept corpus produces answers that are correct and flat. The reflex is to add sources or rework the prompts; the lack is elsewhere, and no further source of the same kind will fill it.
It also gives a criterion for evaluating a workspace: not what it contains, but what it has no means of receiving.
Nuances and limits
The gap is not always costly. On a standard, heavily codified subject, where fine-grained understanding brings nothing decisive, a purely documentary context is enough.
And not everything the product manager has in mind needs to be poured in: part of that material is an impression of the day, which will not survive the week.
Open questions
- How do you spot that a context suffers from this particular lack rather than from poor curation, since both produce the same impression of a superficial answer?