# A/B tests

> Run two versions with the same prospects and get one verdict.

Run the same prospects against two versions, then compare. You get one verdict.

## Run both sides

Label each run with a variant, then compare the newest of each.

```
leakdown https://your-site.com --variant control --yes --headless
leakdown https://new.your-site.com --variant new-pricing --yes --headless
leakdown --compare your-site.com# your-site.com — variants compared

Newest run per variant: `control` → your-site.com/2026-09-17/10-14-02--control; `new-pricing` → your-site.com/2026-09-17/11-30-48--new-pricing.

| variant | sessions | completed | leaked | how sure |
|---|---|---|---|---|
| `new-pricing` | 10 | 3 | 7 | 70% [40–89%] |
| `control` | 10 | 7 | 3 | 30% [11–60%] |

**Verdict:** `new-pricing` leaks more — 40 points more of its visitors gave up than in `control` (40–89% vs 11–60%).

  Written: runs/your-site.com/COMPARE.md

```

`--variant` labels every session and the run folder, so two variants never share a report. `--compare` takes the newest run of each.

## Reading the verdict

For each side: sessions, completed, leaked, an interval, and where it lost people. Then one line:

| Verdict | When |
|---|---|

| Which side leaks more | The intervals do not overlap. |
| No meaningful difference | The intervals overlap. |
| Too few | Under three sessions on a side. |

Note

Sessions that could not run are left out of every rate. Exactly two variants are compared; any more are listed.

---
Source: https://docs.leakdown.dev/ab-tests