You seem to be trying to play decoherence theory against the measurement problem, but the two are actually orthogonal. QM does not make wrong predictions, but we need to be precise about what a "prediction of QM" actually is:
The standard formalism of quantum mechanics gives a prescription for how to calculate the probabilities of the results of measurements of a quantum state (the Born rule), and it gives a prescription for how to calculate the evolution of quantum systems undisturbed by measurement (the Schrödinger equation). It does not, as such, explain what a measurement is or how it works. If we want to be precise, quantum mechanics without an interpretation is not a complete model of "how the world works" (because the Born rule and the Schrödinger equation contradict each other - the evolution induced by the first is not unitary, while the evolution induced by the latter is), but merely a tool that correctly predicts the results of measurements in a laboratory (or anywhere else) by chaining these to different evolutions together in the appropriate manner for the measurement being performed.
You can model the interaction between the measurement apparatus and the measured system as a quantum interaction itself, resulting in a superposition of states of the form $\sum_\lambda c_\lambda \lvert \lambda_s\rangle \otimes \lvert \lambda_a \rangle$ for the eigenvalues $\lambda$ of the quantity being measured, where the $s$ subscript denotes an eigenstate of the system for that quantity and the $a$ subscript denotes the corresponding "pointer state" of the measurement apparatus. This line of thinking began with things like the von Neumann measurement scheme and finally resulted in decoherence theory, explaining how measurement apparatus and measured system end up in this rather peculiar form of superposition. Decoherence does not explain away the measurement problem: It has no explanation for why we see only a single pointer state of the macroscopic measurement apparatus, nor does it claim to. It "merely" explains how the apparatus' state becomes entangled with the measured system in such a way that the pointer state always matches the actual value of the measured quantity for the corresponding state of the measured system.
But whether you model the interaction between the apparatus and the system thusly is irrelevant for the prediction of the probabilities of the results of the measurement (unless your measurement apparatus is imperfect and the $c_\lambda$ are not equal to the coefficients $c'_\lambda$ in the original state $\sum_\lambda c'_\lambda\lvert \lambda_s\rangle$ of the measured system alone) - the Born rule still holds. Adding gravity into the mix is likewise irrelevant - whether we "measure" the extended system of measured system + apparatus by looking at the apparatus it or by somehow deducing the position of the apparatus' pointer from its gravitational field doesn't change anything.
There is no universally accepted answer to the measurement problem. The lack of such an answer is one of the main reasons for the multitude of quantum interpretations, with each interpretation usually offering a different explanation of why we only see a single pointer state.
It is important to note that "the quantum state" of an object is inaccessible to us. All we can observe are the individual pointer states of quasi-classical measurement apparati, we cannot observe an actual quantum state. So whether a system is "in a superposition or not" is not, strictly speaking, a prediction of quantum mechanics. The predictions of quantum mechanics are the probabilities for results of measurements, and nothing else.