Programmatic Page Systems with an Index Gate and Retirement Rule

A page belongs in the index when it can state something true that the other 4,000 pages cannot. We turn that sentence into a gate inside your database: a record enters the sitemap only while it passes a completeness rule, and leaves when it stops passing. Most of this engagement is the argument about where that threshold sits, and the argument is worth more than the templates. The gate: Written as a view your team can query, not as a document. For a used-stock system it might read as four populated attributes, one price, one photograph set and a record touched in the last 30 days. For tariff pages it is a live price with a source timestamp. The count of qualifying records becomes a number marketing can watch move, which changes the internal conversation from how many pages can we generate to how much of our data is publishable, which is the correct conversation. The URL grammar: One pattern per intent, decided against the SERP, with the parameter order fixed and the segment set closed. City plus make. Make plus model plus city. Not all three permutations because a spreadsheet can produce them. Closed segment sets are the reason the system does not quietly grow a long tail of URLs nobody chose. Retirement, which is the half most systems skip: A used unit sells and its page has to die properly. 410 when there is no equivalent, 301 to the model-and-city page when there is, and never a soft 404 landing on an empty results template. Tariff pages that stop being offered are archived with the date they closed rather than deleted, because the query survives the product. The retirement job is specified and scheduled in the same release as the generation job; if you build one without the other you are building an index bloat problem with a 90-day fuse. The build spec: Next.js App Router with ISR and on-demand revalidation fired by your PIM or stock feed webhook, sitemaps sharded at 50,000 with lastmod taken from the record's own updated_at rather than the deploy time, IndexNow pings for Bing and Yandex, and a coverage watch that alerts when the gap between qualifying records and indexed URLs widens past a set point. The page's visible number, its schema and its updated timestamp all read from the same field, because a price in the markup that contradicts the price in the page is a manual action waiting to be filed. Not included: Copy generation. We specify the sentence patterns and the data they read from; we do not ship a language model writing paragraphs, because a generated paragraph is exactly the part of the page that carries no information the record did not already have. We do not build the pages ourselves on the standard engagement, and we do not launch systems like this on a subdomain. Who this is not for: Teams whose data source is a spreadsheet somebody maintains by hand. Teams who want page count as the reported metric. And anyone unwilling to accept that the gate will initially exclude most of their records, which is the intended result and the moment where these projects usually get renegotiated. Where this fits: Vehicle stock and dealer location systems, energy tariff and switching comparisons where the regulated price moves daily, property listings, and marketplace supply pages. Spanish and English, with locale handled as separate systems rather than a translation layer over one.