This work introduces a structure-constrained single-image super-resolution (SISR) framework based on explicit conditioning by a deterministic local structural measure, denoted Φ (phi). Unlike conventional deep learning approaches that encode structural priors implicitly within network weights, the proposed method injects an explicit structural guidance map computed directly from the low-resolution (LR) input. The Φ map is obtained through local sliding-window analysis and is provided to the super-resolution network as an additional conditioning channel alongside luminance information. The central hypothesis is that structurally active regions may benefit from richer high-frequency reconstruction, while homogeneous regions should remain conservative in order to limit unstable artifacts. This principle introduces spatially adaptive and interpretable control over reconstruction behavior. Experiments were conducted on 223 images from the Flickr2K dataset using a ×4 upscaling protocol. Evaluation focuses on structurally selected regions of interest (ROIs) derived from the Φ-based masking pipeline. Quantitative assessment includes PSNR-Y, SSIM-Y, Sobel gradient correlation, LPIPS perceptual distance, and frequency-domain metrics such as FFT high-frequency excess and Laplacian residual energy. Key findings: Φ-conditioned super-resolution achieves the highest median PSNR on structurally selected ROIs. Improvements over classical interpolation baselines (Lanczos and bicubic) are statistically significant (Wilcoxon paired test, n=223). Compared with lightweight CNN and transformer-based models, Φ-SR provides competitive reconstruction fidelity while maintaining moderate high-frequency amplification. Frequency analysis suggests that Φ contributes to a more controlled reconstruction of high-frequency content. This study provides an extrinsic validation of a broader structure-guided image processing framework using standardized super-resolution benchmarks. The work emphasizes interpretability, deterministic structural conditioning, and reproducible evaluation protocols.