For thirty years, traditional and "resolute" readers of the Tractatus have been pressing the same objection against each other in two forms: each establishes that the other can only sustain his reading by borrowing from the one he opposes. Neither camp can account for this symmetry, since each must hold that the other is confused. This paper proposes to see in it a stable structure, not the history of a dispute. A single term, borrowed from mathematics and defined within the text — obstruction — makes it possible to say what that structure is: two complete readings, each in its own context, which do not assemble into one. The paper locates the failure of assembly, draws from it a falsifiable prediction, and identifies the change of viewpoint that makes the finding evident. This record contains the same paper in two versions: English (primary) and French (original). — Ce dépôt contient le même texte en deux versions : anglaise (principale) et française (originale).