The OP describes an aberration between real-time data and archived data, however there could differences in how these two figures are derived.
Real-time data could come straight off the sensor, the BoM like doing this kind of thing for public consumption. Unfortunately sensors could easily be affected by many other external factors e.g precisely dropped steaming bird pooh.
Data for archival would mostly likely be generated from a model that takes as input not just the sensor but many other factors including the dynamics of the weather at the time, some statistical analysis, etc with the aim of providing a more accurate value for the wider area rather than just the immediate area around the sensor bulb.