What gets counted
The current collection contains 20 complete essays, 45,284 words and 1,387 addressable text passages. Headings, paragraphs, list items, captions and nonempty quotations are counted once. A word is a run of English letters or digits, with an internal apostrophe or hyphen allowed. Curly and straight apostrophes are equivalent for search; plurals are not stemmed. Punctuation alone is not a word.
The previous migration’s counter reported 45,308 words using a looser HTML-based rule. Differences from that number reflect tokenization, not abridgement. The original text is preserved and checked against its migration fingerprint. Reading time is an estimate at 230 words per minute, rounded up.
An essay has a shape
Each heading begins a section. The opening before the first heading is “Opening.” Section lengths include their heading and all following passages up to the next heading. The reading band’s widths are proportional to those counts. The adjacent text index is the keyboard and touch alternative. Paragraph addresses are stored in the source, so an unrelated edit does not move every anchor.
A passage beside its measures
The home reading plate quotes one complete original prose passage from the selected essay, with its stable address. It is a mechanical entry point, not a generated summary or a claim about the essay’s central argument. The selector prefers paragraphs of 35–100 words containing at least three of the essay’s recurring eligible terms. It favors up to four distinct terms, then length nearest 65 words, then the earliest passage. If no paragraph qualifies, it considers other prose passages of at least 20 words. Tables, headings, captions and code blocks are excluded.
Up to four words come from the existing frequency-ranked essay vocabulary; numeric-only terms and words appearing fewer than twice are omitted. Highlighting preserves the original spelling and casing. “Here” counts only the quoted passage; the full count includes every measured passage. Section bars compare occurrences of the chosen word and show up to the three most frequent matching sections, with ties in reading order. Their links open the existing reading lens with the exact word and section, at the first matching source passage. The full-count link opens that word across the whole essay. These are recurrence measures, not importance scores. An optional detour uses the same lexical-neighbor model described below.
Shared words, not shared beliefs
Connections compare prose paragraphs, list items and quotations of at least 20 words from different essays. Tables, headings, captions and code remain counted and searchable, but are not suggested as prose neighbors. Four tables flattened by the previous migration have their original rows and columns restored without changing their text or passage IDs.
Common function words and “AI” are excluded. A candidate must share at least two eligible words. Shared words that occur in fewer passages receive more weight: squared log(1 + eligible passage count / passages containing the word), summed and divided by the geometric mean of the two vocabulary sizes. Ties use the essay slug and passage ID. At most three different essays are suggested.
The displayed words are the reasons for a connection. These are lexical neighbors, not semantic similarity scores, fact checks, influence claims, endorsements, or evidence of agreement. A match may be illuminating precisely because the arguments differ. Short passages and passages with no qualifying neighbors say so.
The illustrated collection
The five preoccupations are editorial categories inherited from the original collection. Colored bands identify them. The engraving’s lobes follow the number of essays in a theme; its fine lines and decorative motion are expressive, not measurements. Each tally above the plate represents one hundred words, rounded up. Covers and photographs come from the existing site and can be opened without cropping.
Privacy, dates and unfinished work
Reading and searching happen locally. Only optional reader marks are stored in this browser, with a visible reset. There is no analytics endpoint, live poll or tracking pixel. The research cycle is an authored demonstration, not a running agent system, and gathers no external evidence. The archive’s historical dates came from a sitemap’s last-modified field; unverified publication dates are not presented as publication dates or put into RSS pubDate fields.
The complete site is static HTML. Reading, section navigation, images and the complete essay list work without JavaScript. Search, theme filters and the optional reading lens require JavaScript; no API key or backend is required.