Sales are tested at the start and trusted at the end. That is backwards. Starting a sale is one write per variant and it either works or visibly does not. Ending one is the same write in reverse, except nobody is watching, and the failure looks like nothing at all.
The twenty minute rehearsal
- Pick three products. One with a single variant, one with many, one that already has a compare-at price set. That third one is where most tools get confused.
- Export those products to CSV and keep the file. This is your answer key.
- Schedule a sale on just those three, starting in five minutes and ending twenty minutes later.
- Watch it start. Note how long the write actually takes.
- Look at the storefront, not the admin. The admin shows what was written; the storefront shows what a shopper sees, and those differ when compare-at is wrong.
- Wait for it to end on its own. Do not end it manually, because ending it manually is not the thing you are testing.
- Export again and compare against the answer key, cell by cell.
The whole point
Step seven is the test. Steps one through six are setup. If you stop after watching prices go down, you have rehearsed the easy half.
What the comparison tells you
| What you see | What it means |
|---|---|
| Price restored, compare-at empty | The tool tracked price but not compare-at. Your storefront loses the strikethrough on every future sale. |
| Price restored, compare-at zero | Worse: something wrote a zero. Products will read as full price and no discount will display. |
| Price slightly off, like 41.99 to 41.98 | The restore is recalculating rather than replaying a saved value. That drift compounds every sale. |
| Some products restored, others not | The failure is per variant and silent. Ask what would have told you, had this been 2,000 products instead of 3. |
When to do it
Four weeks out, not four days. A rehearsal that finds a problem in week five leaves time to change tools. The same discovery in the last week leaves you choosing between an untested workaround and no sale.