The net new MRR waterfall

Investors at Cloudpeak do not only ask for MRR, they ask why it moved. Every month the change splits into four movements: new (the first subscription row of an account), expansion (an upgrade: the new row has a higher mrr than the previous row of that account), contraction (a downgrade) and churn (a row whose end_date has no follow-up row for that account). A movement belongs to the month of the start_date of the new row, or of the end_date for churn.

Dataset: Cloudpeak. Every plan change writes a new subscriptions row: the old row gets an end_date and the new row starts on that same date. A row with end_date NULL is the current plan. A customer who cancels has a last row whose end_date has no follow-up row.

Task: for every month from 2026-01 to 2026-06 (generate them with a recursive CTE so quiet months show zeros) return month (YYYY-MM), new_mrr, expansion_mrr (the increase in mrr), contraction_mrr (the decrease, as a positive number), churned_mrr (the lost mrr, as a positive number) and net_new_mrr = new + expansion - contraction - churned. Use 0, not NULL, for empty cells. Sort by month.