Stats

Everything here is counted at build time from the entries themselves, so it goes stale only when the site is rebuilt. Word counts strip markup and count whitespace-separated tokens, which is close enough for a sense of scale and not precise enough to be worth arguing about.

Every entry placed by what it is about, not by tag or date. Two entries sit close together when the language model that powers the site's semantic search puts their text in a similar place; a line is drawn where a pair is genuinely close. Hover a point for its title, or click through.

• Entries are everything in the writing register: blog posts, digesting notes and project notes. Pages like this one are not counted.
• Word counts strip HTML and count whitespace-separated tokens. Code blocks are included, so technical posts read longer than their prose alone.
• Length uses the median rather than the mean, and the range names the shortest and longest entries rather than a standard deviation, which would imply a normal distribution the corpus does not have.
• The two marks in the year table use different scales on purpose. The monthly sparkline is scaled to its own year's busiest month, so it shows the shape of that year: which months were active and how the writing clustered. The bar beside it is scaled across every year, so it shows how the years compare. A tall sparkline in a quiet year does not mean a busy year; the bar is the one to read for that.
• Every number is printed beside its mark, and each sparkline states its own busiest month and active-month count on hover and to assistive tech. See the colophon for how the related-entry suggestions are derived.