# What changed between brief v1 and v2

Fictional GreenLeaf example. AI-assisted editorial review; human release pending.

| Area | v1 | v2 | Evidence class |
|---|---|---|---|
| Mobile signal | Target and authored teaching signal; no checked output yet | Checked 3,780/8,400 → 2,352/8,400 | Reconstructed CSV, verified calculation |
| Audience mix | Compare before diagnosis | Mobile starts shift from equal shares to 25% skeptical / 75% exploring | Reconstructed CSV, verified calculation |
| Comparable groups | Required analytic check | Fixed-mix Week 2 rate 35.33%; all 12 strata decline | Descriptive calculation, no causal interpretation |
| Product history | Unknown; request from owner | Mobile revision reported; audit/assignment/rollout missing | Prewritten fictional context, verification pending |
| Support history | Unknown | Unlinked difficulty finding next-step control reported | Fictional anecdote, no prevalence estimate |
| Next step | Diagnose before a premature intervention | Prioritize mobile/event verification with product/media owners | Editorial decision, proposed not completed |
| AI numbers | Must be checked | Initial/follow-up errors documented; use checked metric pack | Actual recorded responses plus independent checks |
| Business targets and unknowns | Four targets, missing causal/LTV/awareness/economics evidence | Unchanged | Course scenario and evidence limits |

No known history was deliberately removed from v1. The later context was written before the initial
run and represents information learned later by the fictional team. This sequence is not a controlled
test of human superiority, nor proof of a product defect or a measured saving.