nx_search_census.nx

nx_search_census.nx -- G1: THE SEARCH CORPUS, THROUGHPUT AND DIVERSITY CENSUS. /compare/search rung G1, contracts idx_corpus_census and wc_throughput_census. WHY IT EXISTS. On 2026-09-03 the operator could see that the index was not growing and the BOARD COULD NOT. /compare/search carried four latency rows and a relevance row and NOTHING for corpus size, crawl throughput or host diversity; the one row that touched scale had already withheld its Nishi figure as "prose, never a measurement" and nobody had measured it since. Nothing else in the estate measured it either -- knowledge/status/growth_series.log tracks the PM board, debts and maturity and carries no corpus field at all -- so the one axis the operator was complaining about was invisible to every consumer and therefore priced at zero by the ranker. WHAT IT REFUSES TO DO. It prints LEVELS and never a rate. A rate needs two samples in time and the estate ALREADY has a time spine, nx_growth_series, whose report verb computes d/w/m deltas per axis. A second spine here would be the duplicate-ruler defect, so this organ is a PRODUCER for that one: it TRUNCATE-writes a canonical status file and lets the spine carry the trajectory. The throughput axis is the DIFFERENCE between two spine rows, never a number this organ invents. EVERY COUNT CARRIES ITS COVERAGE, and that is not a slogan here. The key census is COMPLETE over the manifest and says so, with its partition summed and the sum printed. The outcome partition is a DECLARED TAIL WINDOW and says so. This organ was written in the same session a sibling validator was caught reading the OLDEST 3.2 percent of a 129 MB log while printing "sampling newest-first" -- a window published without its denominator reads as a population, and that is exactly how a three-week old fossil became evidence about today. ONE FAIRNESS RULER. The fair verdict is wc_frontier_fair, imported from the ENFORCER itself, and the host parser is the crawler's own wc_host. Re-deriving either here would be a second ruler that agrees today and diverges silently on the next edit to one side. DECLARED GAPS, NAMED RATHER THAN HIDDEN. (1) doc= counts seg-store ENTRIES, which are append-only, so it is an UPPER BOUND on distinct documents and the output says so on its own line; distinct-key dedup is the named next rung. (2) The fr: pending-versus-done split is NOT reported: splitting it needs an ss_hget per frontier key and the live shard holds millions, so the honest choices were a silent sample or a named absence, and this takes the absence. usage: nx_search_census [shard-prefix] [outcomes-log] license_tier: ORIGINAL expect_exit: 0