What predicts search impressions across 14 small sites, and one correlation caught decaying in eighteen days

A longitudinal series from one operator's own network: the same 14 sites measured daily on the same 25 variables, 13 measurement days. The stable result: share of pages actually served versus 28-day impressions, median Spearman rho +0.78, never leaving the band +0.69 to +0.89. Article count correlates consistently weaker (+0.37 to +0.56). Publishing more pages is not what moves impressions here; getting the existing pages served is. The interesting result is one that is decaying. Domain rating versus impressions is negative, the opposite of the usual expectation, but it has halved in eighteen days: -0.65, -0.62, -0.49, -0.23. A relationship that decays like that is usually not a relationship, and the composition file shows why: the high-DR sites here are inherited niche domains with few pages served. The correlation measured what kind of site carries which DR in this network, not an effect of DR. It is included precisely because it is the sort of number that gets quoted as a finding. One column is excluded rather than published with a caveat: the collector's age_jours reports 4 for every site, which is wrong. A broken column quoted once outlives the caveat attached to it. Method and limits in the README. Measured for WeThePurple.