nx_kernel_adoption.nx

nx_kernel_adoption.nx -- THE SECOND AXIS: does the feature actually RUN in the live boot? WHY (a measured ruler defect, debt idx 2071): nx_kernel_census computes PRESENT with re_has() substring probes over the CONCATENATED gate logs. A feature proven ONCE by a standalone fixture binary therefore scores IDENTICALLY to a feature wired into the image the pinned BOOTSOV actually boots. Measured 2026-07-31: the census read 382 permil BEFORE and 382 permil AFTER the live boot went from a 72-byte banner exercising ZERO features to a 1480-byte kernel exercising ELEVEN of its eighteen -- the ruler scored the adoption gap as exactly zero movement. A RULER THAT CANNOT DISTINGUISH PROVEN-IN-ISOLATION FROM WIRED-AT-THE-CHOKEPOINT WILL ALWAYS SCORE THE ADOPTION GAP AS ZERO, and this census is the instrument meant to catch it. WHAT THIS DOES: boots the LIVE image in-process (the artifact itself, not a log ABOUT it), captures the serial transcript, and scores every incumbent feature on TWO axes -- evidence : re_has() over the gate logs (what the census already measured) live : its marker appears in the transcript (what the census could not see) Headline = MIN(evidence, live), never the mean and never the max, per the min-not-mean law. ADOPTION GAP = features with evidence but NOT live -- the number the census could not report. Data-driven: the feature->marker table is knowledge/registry/kernel_live_markers.tsv, never hardcoded; a feature with no row there is NOT live, which is the honest default. Self-validating: a positive control must be live, a synthetic marker must NOT be. license_tier: ORIGINAL Shared boot-and-capture: this organ, the gate and the census must agree byte-for-byte on what booting the image means, or the rulers can disagree about the same artifact.