# Control vs rebuild: test plan

**Control (completethyroid.com VSL):** 0.34% CVR, AOV $209.92, 19 sales, $3,989 revenue → ~5,590 visitors, **$0.71 revenue per visitor (RPV)**. (Sales ÷ CVR; the client's visitor count wasn't supplied, confirm.)

**Challengers (each is a different hypothesis, same offer so price is not a variable):**
| Arm | What changes | What stays the same |
|---|---|---|
| A. Control | VSL page | — |
| B. Traditional (TSL) | Long-form mirror-first letter, buy boxes visible from the offer section, compliant claims | 2/3/6 prices, 180-day guarantee, ladder |
| C. Ecom (video top) | Short 7-minute video + product-first page, buy box under the video | same |
| D. Improved VSL (optional) | Improved 19-min script recorded and dropped into the control page with buy boxes at ~10:35 | same |

**Primary metric:** revenue per visitor (protects the strong AOV). Secondary: CVR, AOV, upsell take rate per step, refund rate at 30/60/180 days.
**Do not** call the winner on the page's looks. Call it on RPV once the sample below is reached.

**Sample needed per arm (80% power, α 5%, baseline 0.34% CVR):**
| Lift to detect | Visitors per arm |
|---|---|
| 3× (1.02%) | ~2,300 |
| 2× (0.68%) | ~6,900 |
| 1.5× (0.51%) | ~23,000 |
| 1.25× (0.43%) | ~83,000 |
A three-arm split of ~7k visitors per arm detects a doubling. Smaller lifts need far more traffic; plan on RPV directionally until then.

**Split & hygiene:** randomize at the ad-click level with the same traffic source, same day-parts, run at least 14 days to cover weekly cycles, no page edits mid-test, log the exact page version and deploy ID for each arm. Keep the control running as a concurrent arm (not before/after).

**What we already know is different from the control (so a win isn't confused for luck):** claims removed (no Harvard/25% test, no FDA badge, no ritual/conspiracy story, no fake counters or invented customers); buy boxes not hidden behind a delay; women-first brand and label; honest savings basis (vs the 2-bottle rate, not $159). If the new arms lose, the likely cause is the lost urgency/mystery of the control story, so test one urgency element at a time (real, truthful ones only).