The starting date for the power law is the GB because it makes sense physically. You can also use methods that do not involve regression to demonstrate that this date is consistent with existing data.

But in a power-law process the exact starting date becomes irrelevant at large scales.
The local slope converges to the true scaling exponent regardless of the chosen origin of time.
This is why studying the distribution of local slopes is a much more robust way to test Bitcoin’s power law than fitting regressions anchored to an arbitrary starting point.
The proof is below.