[fidget]

No one does, you estimate each task to have a different number of points, which modify the task's contribution to the burndown graph.

I guess the idea is that points should correlate to time (questionable), and if you're estimating perfectly, that should give you a linear burndown graph. Which is stupid. What might be less stupid is that if you want to get some idea of how well your team is estimating tasks, compared to i.e. this time last year, you could look at the volatility (across several sprints) this year compared to last year, and a lower burndown rate volatility might suggest that the team has improved at estimating the duration of tasks.

However, when you just focus on the volatility as a metric, then people start optimizing for it, which is not the point at all.