Main idea

In desk research, "I've searched enough" isn't decided by a number — 30, 50 or 100 sources are not an epistemic criterion. The convergence signal lies elsewhere: new iterations no longer create core knowledge, no longer significantly change the important notes, no longer reveal any major contradiction or gap in the scope, mostly add confirmation, and leave the remaining uncertainties clearly identified.

That signal can be put to the test. You deliberately run searches designed to break the model — to find what would force you to revisit a central conclusion. If they fail several times in a row, saturation becomes credible. In preparing for Adobe Summit Paris 2026, iterations 13, 14 and 15 served that purpose, and the fifteenth retained no source important enough to change the model of the five themes.

Convergence established this way doesn't say "we know everything". It says that the probability of a new iteration substantially changing the representation has become low relative to its cost.

Why it matters

Saturation observed passively — nothing new turns up any more — may come from research that is no longer looking in the right place. Saturation reached after failed attempts to break the model is a result, not weariness.

And the criterion is economic as much as epistemic: it weighs the chance of changing the model against the price of one more iteration.

Nuances and limits

Break searches are designed by whoever holds the model. They target what that person can imagine as a threat, and can miss an objection coming from a frame they don't know.

And convergence remains relative to the scope: the subject isn't exhausted, only the mission is.

Open questions

• Should break searches be designed by another agent, or another person, than the one who built the model?