Much of this is outdated and wrong. Some of it can not even be done on modern kernels. If nothing else please do not apply the advise given on tcp_rmem defaults. Also do not set large MTU on home internet facing connections as this will break many internet facing connections. Same for tcp fastopen, as this is handled by QUIC and not meant to be set on all tcp connections as the security around this was intended for use in the datacenter. The low fin timeout should only be applied on busy web servers that are not behind load balancers and only if those involved know what they are doing. The limit output bytes is also intended for web servers using http/2.0 and should not be set universally. Window scaling should already be enabled on nearly all kernels in use today. The 16MB initial window size will certainly break connections to some load balancers, middle boxes and some security focused firewalls. There is another setting that should already be on that negates some of this thankfully. Someone will test this and say it works fine from their workstation to some VM or server but the internet is much more diverse than that.
I am not going into details on all of this as it would at best turn into a massive blog. For most people the defaults should suffice and for others that need to improve latency they should learn not only what each setting does but how they play together which also involves learning a great deal about real world implementations of TCP/IP. For those that love to tune things, myself included, tune one thing at a time and give it a month or two to soak whilst you test communication to every possible thing. Don't just use tools like iperf, nuttcp and other benchmarking tools. They will deceive even the smartest people as they do not mimic the real internet. Behaviors change dramatically as latency, network appliances, load characteristics, packet loss, packet reordering and other things get in the way. If going down the path of tweaking things, first get very comfortable with tcpdump and Wireshark.