Melt inclusions record the depth of magmatic processes, magma degassing paths, and volatile budgets of magmas. Extracting this information is a major challenge. It requires determining melt volatile contents at the time of entrapment when working with melt inclusions that have suffered post-entrapment modifications. Several processes decrease internal melt inclusion pressure, resulting in nucleation and growth of a vapor bubble and, time permitting, diffusion of volatiles (especially CO2) into the vapor bubble. Methods exist that attempt to reconstruct the entrapped CO2 contents, but they are difficult to apply and yield inconsistent results. Here, we explore bubble growth, evaluate CO2 reconstruction approaches, and develop improved experimental and computational approaches. Piston-cylinder experiments were conducted on olivine-hosted melt inclusions from Seguam (Alaska, USA) and Fuego (Guatemala) volcanoes at the following conditions: 500-800 MPa, 1140-1200 °C for Seguam and 1110-1140 °C for Fuego, 4-8 wt% H2O in the KBr brine, and run durations of 10-120 minutes. Heated melt inclusions form well-defined S-CO2 trends that can be described by degassing models. CO2 contents are enriched by a factor of ~2.5, on average, relative to those of the glasses within unheated melt inclusions, whereas S contents of heated and unheated melt inclusion glasses overlap, indicating insignificant amounts of S partition into the vapor bubble. Low closure temperatures enable CO2 diffusion into vapor bubbles during quench upon eruption, while a higher closure temperature for S limits its loss to vapor bubbles. We evaluate the timescales of post-entrapment processes and use the results to develop a new computational model to restore entrapped CO2 contents: MIMiC (Melt Inclusion Modification Corrections). Heated melt inclusion data are used as a benchmark to evaluate of the results from MIMiC and other published methods of CO2 reconstruction. The methods perform variably well. Key advantages to our experimental rehomogenization technique are that it enables accurate measurements of CO2 contents and allows for large quantities of melt inclusions to be rehomogenized efficiently. Our new computational model produces more accurate results than other computational methods, has similar accuracy to the Raman method of CO2 reconstruction in cases where Raman can be applied (i.e., no C-bearing phases in bubble), and can be applied to the vast body of published melt inclusion data. To obtain the most robust data on bubble-bearing melt inclusions, we recommend taking both experimental- and MIMiC-based approaches.