AI 替代下的验证不对称：工资比反转与学徒培养阈值

This paper develops a two-sided theory of verification asymmetry under AI substitution. When verification skill is acquired through the work later evaluated, AI can raise the relative marginal value of a verification stock while reducing the experience flow that reproduces it. A CES production block yields a closed-form crossing of any reachable verification-to-generation marginal-product ratio at fixed verification stock. A cohort block makes supply endogenous: juniors accumulate generation experience at rate 1-theta, and promotion depends on that experience. Hard promotion produces a discontinuity at theta*=1-tau*/T_j; smooth promotion replaces the jump with a kink and raises the stock exponent from a to a+b. After a permanent threshold-crossing step, exact cohort accounting separates already-trained retirement from a later full-stock zero guarantee. Under stationary competitive prices and incomplete employer capture, the cohort mechanism yields a discounted apprenticeship-value residual. Exact Cobb-Douglas aggregation propagates a zero component, whereas every fixed elasticity sigma_a>1 preserves a positive residual that converges to zero as sigma_a approaches 1. Together, the results define a portable mechanism for task systems where AI expands generation and verification capacity is reproduced through experience, yielding targets for price-stock divergence, cohort thresholds, and employer capture.