Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Angle
A prior-art search finds nothing as long as it is conducted from the place where the idea was born. The author's vocabulary has been used by nobody, his discipline has treated only one version of the problem, and his language indexes only one corpus among others — three invisible filters producing the same reassuring result: no antecedent. Each is lifted by a shift, and each shift brings down part of what you thought you held. What remains at the end is not a new idea but a composition, and a composition is a sturdier answer than a flag, because it survives the discovery of the next antecedent.
Synthesis
The three shifts are not interchangeable: they follow on from one another, and each is only possible once the previous one is done. Leaving your vocabulary presupposes having broken the method down into components and relations, because you can only question objects small enough for an answer to exist. Leaving your discipline presupposes asking the questions by function — how do you pick up interrupted work, how do you hand a case over — and not by concept. Leaving your language presupposes replaying those same questions elsewhere, which brings to light entire traditions that English did not index.
Two rules hold the exercise together, and they look in opposite directions. The first commands the search: look for the best counter-example rather than for reasons to believe, and revise the hypothesis every time a counter-example appears. The second closes it: a new source only counts if it changes the map — older, markedly closer, or such as to change the conclusion. Without the first, you gather confirmations; without the second, you carry on until you manufacture the resemblances you came looking for.
The hierarchy of evidence follows the same logic: a system that has run shrinks the claimable territory more than a conceptual article from the same family, because it paid the price of the trade-offs. And the final judgment remains a judgment — close enough to be interesting is not the same thing, and both sentences have to be holdable together.
Tensions / contradictions
The two rules pull against each other. Looking for the best counter-example pushes you to carry on indefinitely, since one more counter-example is always possible; requiring a source to change the map pushes you to stop early, on a map that is the one you already have. The stopping criterion is therefore judged with the very instrument it is meant to control, and a search that started badly saturates quickly with nothing to signal it.
A second tension sets the claimed product against the use made of it. The search concludes that a verdict of originality is undecidable and that it is better to describe a composition — but a patent filing, a challenge or a claim of authorship demand precisely the verdict the method refuses to produce.
Questions
- Who can spot the tipping point towards the manufacture of resemblances, when the person who has been searching for weeks is the last to see it?
- How do you reach the systems that were built and never published, which are the most decisive and least accessible antecedents?
- Does a composition map keep itself up to date as the method evolves, or does it remain the snapshot of a moment?