Randomness describes one inherent property of self-assembled metamaterials, and greatly limits the practical applications of metamaterials based on Bottom-Up techniques, such as the micro-sphere lithography technique. Herein, by subtly utilizing the randomness in long-range disorder metasurfaces, we demonstrate a high-performance one-shot full-Stokes polarimeter in the visible waveband. The long-range disorder metasurfaces, i.e., chiral shells, were realized by depositing Ag on the self-assembled micro-sphere monolayer comprised of many micro-domains of random lattice directions and areas. The distinct optical anisotropy and chirality in different micro-domains can result in distinct photo-currents to the photodetector array placed underneath upon the injection of polarized lights. Through establishing the mapping relationship S^=fI^ between the detected photo-currents I^ and the states of polarization (SoP) S^ with the convolutional neural network (CNN) algorithm, we realize a highly-precise full Stokes polarimeter in the waveband ranging from 500 nm to 650 nm, and the minimum mean square errors (MSE) can reach to about 0.37% (S1), 0.33% (S2) and 0.19% (S3) at 566 nm. The averaged MSEs in the investigated waveband are 0.49% (S1), 0.45% (S2) and 0.31% (S3), respectively. The macro- and micro- optical properties of chiral shells, the optical randomness of chiral shells in different domains, the reference SoP number, the exposure time and pixel number of CCD, as well as the reliability and stability of system have been investigated systematically.