https://nosbin.com/nevent1qqs0rhv4frd3hxtu290hf7f803k0ulhq66e3k8p822l8wr2upm799zgpzemhxue69uhkzarvv9ejumn0wd68ytnvv9hxgqg4waehxw309ajkgetw9ehx7um5wghxcctwvsq3wamnwvaz7tmwdaehgu3wvekhgtnhd9azucnf0gq3gamnwvaz7tmwdaehgu3wdau8gu3wv3jhvqgswaehxw309ahx7um5wgh8w6twv5q3jamnwvaz7tmwdaehgu3w0fjkyetyv4jjucmvda6kgqghwaehxw309aex2mrp0yhxxatjwfjkuapwveukjqg5waehxw309aex2mrp0yhxgctdw4eju6t0qyt8wumn8ghj7un9d3shjtnwdaeky6tw9e3k7mgprfmhxue69uhhyetvv9ujummjv9hxwetsd9kxctnyv4mqzxrhwden5te0wfjkccte9eekummjwsh8xmmrd9skc84hp7n

it seems gzip compressed binary data takes about half the size of what gzipped .content of notes take on average.

500 million notes should fit in 25 GB

still need to keep some reference to event ids to find them, linking row number with event id, but this does not need to be in ram while searching.