random_page_cost
The planner's estimate of what it costs to fetch a page from a random location. The default of 4.0 describes a spinning hard disk from the 1990s.
PostgreSQL's cost model is relative: a sequentially read page costs 1.0, and the four means a random fetch is assumed to be four times more expensive — which held when the difference was a physical arm that had to move. On SSDs and NVMe there is no such difference, and the ratio sits closer to 1.1. Leave the value untouched and the planner systematically overvalues index scans, choosing to sequentially scan whole tables where an index would be faster — one of the commonest findings when somebody asks why a query suddenly takes twenty seconds. Adjust per tablespace if you mix disk types.