3D Gaussian Splatting (3DGS) has achieved photorealistic novel view synthesis, yet its adoption is hindered by two challenges: reconstruction sensitivity to unstable input frames and the lack of robust model identification, as conventional watermarking compromises visual fidelity. This paper introduces SplatID, a framework addressing both issues. First, we propose a multi-stage frame filtering pipeline that prunes low-quality frames by leveraging optical flow, SIFT-based geometric validation with RANSAC, and photometric consistency checks. Second, for copyright protection, we introduce a non-perturbative geometric descriptor for 3DGS models. Our method generates a unique signature by identifying salient keypoints via local curvature estimation and encoding the statistical moments of their spatial distribution into a compact hexadecimal hash. This efficient, CPU-based process enables near real-time model identification. Experiments show our filtering significantly improves reconstruction fidelity (PSNR, SSIM), while the hashing mechanism outperforms traditional watermarking in speed and robustness without any visual degradation. SplatID provides a practical toolkit for enhancing 3DGS data quality and protecting the resulting assets.