P5-2 dependence audit (v13)

The dependence-aware companion to the P5-2 story: a stratified bootstrap over the forty persona × conflict-cell clusters preserves the empirical point (0.128) with interval [0.071, 0.189], while the fixed-panel Jeffreys aggregation gives median 0.172, [0.152, 0.195]. Both sit below the frozen 0.20 boundary — but the subsequent prior sweep showed that Bayesian proximity to the boundary moves with α, which is why the bootstrap is the principal sensitivity.