I've always had strangely accurate time estimates. The basic idea is to divide up the work, and make a fair estimate each part, with special attention to unknowns (which get much more time).

Then double everything.

The individual task estimates are often out, but the overall estimate is close... as if, with a population of tasks, there's regression to an accurately estimated mean.

(An alternative explanation is that I over-estimate, and Parkinsonianly, work expands to fill time.)