Copy–move forgery remains one of the most prevalent manipulation types due to its visual consistency and strong concealment. This study introduces FORGE-KEY, a forensic-guided framework that enhances localization by integrating residual cues, frequency-domain patterns, and geometric consistency. A forensic evidence map is first constructed from residual magnitude and resampling periodicity to guide keypoint activation in low-texture and heavily processed regions. A compact Phase–Zernike–Residual descriptor then jointly encodes phase structure, rotation- and scale-invariant geometry, and residual texture irregularities. Matching is refined using residual filtering and multi-model geometric verification to suppress false correspondences. Experiments on FAU, CoMoFoD, CASIAv2, GRIP, and COVERAGE show consistently strong performance, with the highest F1-scores on FAU, CoMoFoD, and GRIP, and competitive accuracy on the remaining datasets. The framework maintains efficient runtime while improving robustness under compression, noise, and geometric distortions.