AI Transformation Experiment

How It Works (Plain English)Start here. No background needed — what a claim is, what a seam is, and how one round of the two-team race runs, explained like a game of telephone and a relay race. Is It Fair? (Plain English)How we keep the simulation honest and grounded — the six safeguards against rigging, the real-company data it's built on, and the awkward result that proves it fights back. Findings ReportThe full arc: theory, instrument, two seeds of data, and the adversarial audit that found the simulation was rigged toward the answer I wanted. Experiment OverviewA controlled simulation comparing organic AI discovery against traditional change management inside Progressive Insurance — 16 sprints, two tracks, one question. Track A — SWT Framework AlignmentHow each SWT principle maps to a concrete simulation component, the 8 concerns we measure, and the 6 gaps we operationalized. Track B — Traditional Change ManagementKotter's 8 Steps, Prosci ADKAR, McKinsey 7S — the standard enterprise playbook, executed well. What it measures, what it focuses on, what it does not do. Measurement, Validation & Proof FrameworkHow we capture results, measure transformation across 6 dimensions, audit every output, and prove the simulation is valid. Workflow ArchitectureThe three simulated business workflows — auto claims, commercial underwriting, and subrogation — how they connect, where AI inserts, and the seams where translation debt accumulates. Results DashboardLive results from the latest simulation run — metric divergence, discovery engine outputs, THEARI phase progression, and agent behavior breakdown. Cross-Run ResultsEvery run side by side — which outcome patterns hold across seeds and which are per-run noise. The "robust, not fragile" evidence. Revision LogThe four rounds of fixing the model itself — what each sweep showed, the mechanism behind it, and what a better Track A and Track B each teach us.