CEACAM5 is a known glycoprotein antigen for which reliable measurement relies on repeatable concentration tracking through repeated measurements. This study proposes a label-free optical sensor based on the CEACAM5-induced modal redistribution analysis of tapered multimode microfibers in terms of their residual optical signature as opposed to the label-dependent identification of fixed images. The problem posed here concerns whether the CEACAM5 concentrations in the range of 1 ng mL−1–1000 ng mL−1 can be determined from the residual speckle signature with the retention of the initial steepness in the lower concentration part and the subsequent saturation at higher antigen concentrations. The data used include 900 specklegrams obtained at 9 different concentrations with respect to a tapered waist of 7.46 μm, tapering length of 15000 μm, near-infrared illumination wavelength of 1550 nm, 6 milliseconds exposure, and separate calibration, validation, and independent testing sets of the acquired images. In each case, a residual optical field with respect to the reference state is determined and characterized in terms of entropy, radial Fourier energy, gray-level co-occurrence texture analysis, angular anisotropy, lacunarity, and wavelet energy. The residual fields are used as descriptors to establish a CEACAM5 concentration path along which branch-aware Gaussian processes calibration takes place. For the lower concentrations between 1 ng mL−1 and 50 ng mL−1, a slope of −0.00123 response units per ng mL−1 with a coefficient of determination of 0.99221 is achieved, while the saturating trend across the whole interval is described with a lower response constant near 0.82648 with R2 = 0.99173. The findings answer the research question affirmatively: residual speckle dose-path calibration converts CEACAM5-induced modal redistribution into an uncertainty-qualified concentration estimate without reducing the assay to categorical image recognition.