EA-MEDIATED-SHARE-01 is a registered forecast on how much economic decision-making AI systems mediate, issued on 2026-09-20 so that its figures can later be checked against named sources. It distinguishes three overlapping events — a buyer consulting an AI system, a session referred from an AI surface, and a transaction completed by an agent — and insists that figures counting different events cannot be summed. Against the US Census retail base it shows that the McKinsey and Bain 2030 forecasts differ by a factor of 1.8 to 3.3 and cannot be compared, because only one defines its scope. It estimates AI-referred dollar share at 3 to 8 percent by end of 2026 under an equal-order-value assumption it labels unverified, and agent-executed share at 5 to 15 percent by 2030. Its valuation is a curve rather than a number: value equals encounters times the measured change in entity-resolution accuracy times an unmeasured probability of outcome change times margin, with that probability left as an axis. It specifies a matched, randomised trial against conventional structured data, and argues that no public corpus other than the Capture Registry holds dated, transcript-preserved observations of entity resolution across surfaces.