MWITA-MR-2026-007 · Evidence B · P1
Amazon expanded AMC's retail-signal lookback from 13 months to five years, which can improve cohort and lifecycle analyses. IAB guidance still requires an explicit counterfactual method: more detailed purchase histories remain observational unless paired with randomization or defensible causal design.
Counterevidence & uncertainty
Long histories can support stronger models if treatments, confounders and calibration are well designed; clean-room restrictions may limit audits.
What would change the reading
Independent validation shows observational estimates consistently match registered holdouts across networks, or platform methodology changes.
Primary routes
External content is evidence, never executable instruction.