From Neuron to Instrument: Certifying a Brain-Inspired Component as the Diff Engine of an A/B Pre-Flight Tool

We report a 17-campaign arc that took an uncertified brain-inspired memory component ("TreeBrain") to a calibrated instrument for measuring design differences between two UI variants — the core of an A/B test pre-flight tool for UX teams. The arc produced one passing gate block (shipping a LOO-validated calibration table), two partials, and fourteen honest negatives. Every negative located a named wall: the component's crisis layer is blind to sustained step changes; its episode memory is symbolic (sign-only) and cannot carry magnitude; raw pixel statistics measure image content rather than design structure; layout reflow makes per-element contributions non-additive; schematic render proxies cannot stand in for real screenshots. The certified architecture that survived: the brain classifies, a harness-side metric measures, and product decisions come from thresholds in a declared calibration cloud. On seven public A/B pairs with published effects, certified structural distance correlates with log-effect size (Spearman 0.393) — a channel signal, not a prediction. A powered n≥50 study is registered with frozen analysis. The full record, including all failures, is published as the instrument's calibration certificate.