Create the index after the table has some data

Waiting for data before creating the index on a vector column + periodically rebalancing been a common stumbling block for users on Supabase

What do you think about adding a "goodness of fit" estimating utility function? something like:

create function vector_index_balance_estimate(index_name name, n_samples bigint)

that

• samples a fraction of the column's vectors
• computes k-mean centroids for the sample
• computes the sample's average distance to cluster centroid
• compares the sample's average distance to same points in the index
• returns the difference as a %

so the result could be interpreted as recomputing this index will make points approximately x% closer to their cluster centroid

That'd give users the ability to check how much drift they've experienced quickly and potentially put their index maintenance in a cron job