You can easily construct examples where the torsion is not of constant type, so the next step in the method is not defined. So I imagine the question is whether you might have termination under the assumption of constant torsion type at each step. But that is somehow unnatural, because we can't really figure out how to enforce such torsion hypotheses directly in terms of a given G-structure, say in coordinates for example, without actually carrying out the prolongations. But even under such strong hypotheses, there is no proof yet, even in the analytic category. I don't see how it could be a consequence of Cartan-Kuranishi. For structures which are flat to high enough order (to acheive involution), there are results of Pierre Molino which imply flatness in the smooth category.

You can easily construct examples where the torsion is not of constant type, so the next step in the method is not defined. So I imagine the question is whether you might have termination under the assumption of constant torsion type at each step. But that is somehow unnatural, because we can't really figure out how to enforce such torsion hypotheses directly in terms of a given G-structure, say in coordinates for example, without actually carrying out the prolongations. But even under such strong hypotheses, there is no proof yet, even in the analytic category. I don't see how it could be a consequence of Cartan-Kuranishi. For structures which are flat to high enough order (to acheive involution), there are results of Pierre Molino which imply flatness in the smooth category.

Edit: Sorry, I forgot that there is a paper of Bernard Malgrange, Sur le problème d’équivalence de Cartan, Annales de le Faculté des Sciences de Toulouse, Tome XXVI, no5 (2017), p. 1087-1136, which proves that Cartan's method terminates in finitely many steps, when carried out on an algebraic pseudo-group on a complex algebraic variety. The proof uses Malgrange's proof of involutivity (after sufficiently many prolongations, above the generic point) of analytic exterior differential systems: Bernard Malgrange, Systèmes différentiels involutifs, Panoramas et Synthèses, vol. 19, Société Mathématique de France, 2005, vi+106 pages. I am embarrassed to admit: I have not read either paper.

You can easily construct examples where the torsion is not of constant type, so the next step in the method is not defined. So I imagine the question is whether you might have termination under the assumption of constant torsion type at each step. But that is somehow unnatural, because we can't really figure out how to enforce such torsion hypotheses directly in terms of a given G-structure, say in coordinates for example, without actually carrying out the prolongations. But even under such strong hypotheses, I don't think there is any proof. I don't see how it could be a consequence of Cartan-Kuranishi.

You can easily construct examples where the torsion is not of constant type, so the next step in the method is not defined. So I imagine the question is whether you might have termination under the assumption of constant torsion type at each step. But that is somehow unnatural, because we can't really figure out how to enforce such torsion hypotheses directly in terms of a given G-structure, say in coordinates for example, without actually carrying out the prolongations. But even under such strong hypotheses, there is no proof yet, even in the analytic category. I don't see how it could be a consequence of Cartan-Kuranishi. For structures which are flat to high enough order (to acheive involution), there are results of Pierre Molino which imply flatness in the smooth category.