Search Console said delete nothing: 428 generated pages indexed, 2,290 handwritten ones never fetched

The two buckets in that report are different verbs, and the summary count hides it. "Crawled, currently not indexed" is a decision Google reached after spending a request on the page ^[11]. "Discovered, currently not indexed" means the URL is known and no request has been spent on it yet ^[12]. The timezone family is in the first case and Google is finished with it: essentially all 1,432 fetched, 428 kept, 647 thrown out, almost nothing left queued ^[8]. The 2,290 recipes are in the second, all of them ^[13]. One is a rating. The other is a place in a line.

That distinction is what turns the cleanup plan into damage. Of the 638 indexed URLs the author actually exported, 428 are timezone pairs, so the collapse-and-redirect would have retired 67 percent of everything Google currently shows for the domain ^[17]. The recipes, which are about 39 percent of the site by page count ^[22], would still be unfetched the next morning, because nothing in the timezone directory was holding them back. The Links report reads external links: 0 ^[14].

The arithmetic deserves one caution the author raises himself. Keep 428 and reject 647 and you have accounted for 1,075 of 1,432 pages; 357 are missing from the tally ^[20]. He notes that the indexed column is a complete list while the crawled and discovered columns come from drilldowns that cap out near a thousand example URLs ^[16]. So the 30 percent acceptance rate is a census divided by a family size, which holds, but the 647 rejections underneath it are a sample floor, not a count. The gap is either sampling loss or pages that were never crawled at all, and those imply different next moves.

The reading hazard in his method is worth more attention than the script. His first pass at the crawled-not-indexed drilldown returned 660 rows, of which 638 were the indexed list still sitting in the reused DOM ^[10]. That is 97 percent contamination ^[21], and it fails in the most expensive direction: it renders as "Google crawled my generated pages and rejected almost all of them," which is exactly the theory he walked in with ^[5]. A quick eyeball of that table confirms the wrong diagnosis and ships the redirect.

What the four minutes of exporting bought was not a verdict on content. It was the ability to tell a quality rejection apart from a queue position, and only one of those two responds to editing pages ^[15]. The two generated families hold 502 of the 638 indexed URLs and the site's two best acceptance rates ^[7], which leaves 136 indexed pages across the tools, hubs and cooking guides ^[19] and none at all from the recipes ^[9]. On the evidence, the handwritten work has not been judged yet. It cannot be, until something outside the domain points at it.