I'm consistently wrong about how long tasks take. How did you actually get better at estimating?

Six months in and my estimates are a running joke to myself. I say "half a day" and it's two days because of some edge case, or a flaky test, or a review round I forgot to count.

The senior folks on my team somehow give a number and mostly hit it. Is that just pattern-matching from years of scars, or is there an actual method? Do you pad by 2x? Break tasks smaller? I'd love to stop dreading the "so when will this be done" question in standup.