Log
A running record of the open-source work. Written when it lands, not reconstructed later.
Rebuilt the site below the hero: an open-source catalogue led by the flagship three, a research section, this log, a labs section and a sponsor page. The vendored copy of the old catalogue is gone; the source of truth is velofy/open-source.
The escalation tier is no longer tied to one vendor. A page that scores below threshold can now route to whichever provider is configured, which keeps the deterministic-first design intact while removing a single point of dependency.
Added ways out for the cases where an index is the wrong tool: forcing a live scan, bypassing a stale index, and opting a path out entirely. An index that cannot be overridden is a cache that lies.
Both docs sites moved onto a flat Geist and amber system, dropping the cyan and purple. The two projects argue for the same thing, so they should not look like they came from different labs.
Encoding detection, conditional cache revalidation, robots.txt crawl-delay, gzipped sitemaps, and proxy rotation on a block. All of it unglamorous, all of it what maintaining a scraper actually consists of.