The challenging environment of in situ micro-geometry measurements in fluids (e.g. for laser- or electrochemical machining), such as refractive index fluctuations, small dimensions and high surface gradients, hinder many conventional measurement techniques. Confocal microscopy remains most suitable with uncertainties < 1μm, but complex micro-geometries with edge slopes > 75° often produce unwanted artifacts. To prevent the formation of artifacts, the isotropically emitting fluorescence light of a fluid layer covering the specimen is measured instead. The geometry reconstruction for in situ-relevant fluid depths >100 μm is not trivial and requires a signal model that includes the contributions of light absorption and the confocal volume shape. For model validation, the surface position of a reference step-object (nominal height: 250 μm), submerged in a fluid layer > 1 mm, is determined using a fluorescence signal model that is fitted to the measured data. First experiments yield a step height uncertainty of 8.8 μm, about one order of magnitude above the requirement. In order to identify optimization potential, the minimum achievable measurement uncertainty is estimated for both a signal with experiment-equivalent variance and a shot noise limited signal. The estimated uncertainties are 3.8 μm and 0.1 μm, respectively, and decrease with lower fluorophore concentration and fluid thickness. The differing experimental and estimated uncertainties result from model simplifications such as the missing contribution of reflected light at the specimen surface where the current model assumes that the confocal volume is cut off. Expanding the model promises to reduce the measurement uncertainty and to converge estimation and experiment, enabling geometry measurements of complex micro-geometries with different surface reflectivities under challenging in situ conditions in fluids.