mirror of
https://github.com/zooai/ngo.git
synced 2026-10-30 15:04:49 +00:00
zeekay
858a193e22
The numbers were typed by hand and three of the four were wrong: 86 papers when there are 88, 149 proposals when there are 146, and 45 models against 78 — the site published barely half its own work, and a `45+` beside it hedging a number nobody could check. Counting at build rather than in getStaticProps, because that spends the whole sixty-an-hour allowance on a single build: Next runs nine workers and then exports every page a second time, so nothing a module remembers survives. One script writes the counts before the build and the whole site compiles with them, which also means every page is right with JavaScript off rather than the ten that had been plumbed by hand. Three requests, one build. A browser then counts at most once an hour and remembers what it found. Without that a visitor spends two of their sixty requests on every page they open — thirty pages before the sources refuse, and the refusal is invisible because the page simply keeps the number it already has. Nothing renders a zero. A source that fails keeps the last number that answered, a truncated tree throws rather than reading as a short count, and a value that is not a positive integer is refused. Papers are the directories holding their own .tex, which is the convention the repo already follows, so a paper added tomorrow counts itself and nothing here keeps a list. proofs stays typed by hand: that repository is private, so a public site has nothing to count.