A table is the most semantic thing in a content page. Everything else on Nakodo 's public pages is prose, lists and drawings; the tables are where a reader goes to compare two things, which means they are also where markup mistakes cost the most. I had never audited them, so I fetched all 76 pages in the sitemap and counted.
Method
Parse each page with Python's html.parser rather than a regex. That matters here: every page carries a React flight payload inside a <script> , and that payload holds an escaped copy of the markup. A regex for <table counts both copies and tells you the site has twice as many tables as it does. html.parser treats script content as raw data, so the count is of real elements.
for f in sorted ( glob . glob ( "/tmp/nak_html/*.html" ) ) : p = TableCensus ( ) p . feed ( open ( f , encoding = "utf-8" , errors = "replace" ) .…