So, forgetting an index and then growth pushing you over the threshold is a valid concern. I think every dev has run into that at some early point in their care

So, forgetting an index and then growth pushing you over the threshold is a valid concern. I think every dev has run into that at some early point in their career. But what your comment is skipping past is there's potentially a 100x or higher difference in throughput for sequential scans vs indexes. If you know the data will have a bounded size this means indexes aren't necessarily a good choice. SSDs have narrowed this gap a great deal but it still exists because of latency and unpredictable prefecting. It even exists with pure in memory applications. Another key aspect is how much of the data the query ends up touching. If you're hitting 25% of the data anyhow a linear scan is likely faster. There's also more niche ideas like arranging to convoy multiple queries along one linear scan, something impossible with indexed scans. It's useful to know about this asymmetry and that sometimes a brute force linear scan is in fact the best tool.