Expected Value Method (EVM)
For symmetric uncertainties the catalog uses the Expected Value Method: each measurement is a normal distribution, the distributions are summed, and the mean of the sum is the estimate. Its uncertainty is the larger of the internal error (from the quoted uncertainties) and the external error (from the scatter of the values), so an underestimated error bar cannot dominate the result. See Birch & Singh (2014) and Singh & Birch (2016).
Weighted mean
The object pages also show the conventional weighted mean with weights 1/σ². A measurement with an underestimated uncertainty can dominate it; comparing both averages shows when that happens.
Asymmetric errors
Lower and upper uncertainties are stored and shown separately. When several estimates are combined, each likelihood is approximated with the lower or upper error on either side of its value; the combined estimate gets separate bounds, widened when the measurements disagree more than their errors suggest. Density errors are propagated separately to first order through ρ = 6M / (πD³); the larger bound sets the summary error and the rank.