Six of 500 programmatic pages drew an impression in the build's first 15 days

Impressions count query matches. They are a lower bound on indexing: a page can be in the index and never be served. The file reports clicks, impressions, CTR and average position from Search Console ^[3]. It does not include an index coverage count, so six pages with impressions ^[9] tells you how many URLs have won a query match and leaves the status of the other 494 open. The post reads that as a queue: "The other 494 pages aren't failing; they're queueing" ^[13], because "Google evaluates a fresh long tail for months, not weeks" ^[14].

Across 500 URLs, 89 impressions works out at 0.18 impressions per page for the fortnight ^[21]. The distribution matters more than the mean here. One page, traditional-bengali-bridal-kolkata, took 64 impressions with no clicks and a position the team says is trending up ^[10]. The remaining 25 impressions are spread across the other five pages that surfaced, roughly five each ^[22].

The 6.7% CTR is six clicks. A seventh click on the same 89 impressions would print 7.9%, a fifth would print 5.6%, so the published rate moves about 1.1 points per click ^[23]. The post calls 6.7% a good number and says roughly one searcher in 15 clicked where Google surfaced these pages ^[18]. For that rate to transfer to another build, the query mix would have to look like this one: long-tail local searches naming a neighbourhood, a tradition or a price band, where a specific page title answers the search better than a homepage ^[5]. It would also need enough clicks to be a rate rather than a count.

Search Console's "last 3 months" filter on this property held 15 days of data, because the property was two weeks old ^[12]. A filter labelled three months that contains a fortnight is the sort of thing that survives into a client deck. The team said presenting it that way would be "the dishonest kind of framing", and named the honest window as two weeks ^[12].

Two inputs are gone by the team's own account: the candidate count before the publish gate, and the gate's rejection totals ^[15]. They declined to reconstruct them, writing that "estimates dressed as data are exactly what this industry doesn't need more of" ^[16]. From the next project, called Glasshouse, rejections are to be counted and published weekly ^[15]. The post also begins to describe a defect the data surfaced in its Pages report, and the published text ends mid-sentence ^[20].

The gate required three things of every page that shipped: a distinct demand the team could name, a data backbone specific to that page, and an intent no sibling page had already claimed ^[6]. Rank tracking covers 432 keywords against 500 live pages ^[8]. If one distinct intent per page is what the gate enforced, at most 86% of those intents have a tracked position ^[25].