# Step 4: buy upgrades

🌐 **English** · [Dansk](Step-4-Buy-Upgrades-da) · [Bahasa Melayu](Step-4-Buy-Upgrades-ms)

`get_store_items` reads the 20 store products, `choose_upgrade` picks the most expensive one the bot can afford, and `buy_upgrade` clicks it. The screenshot shows the first cursor bought.

**Screenshot**

![Step 4: buy upgrades](https://git.tirsystem.com/Tirsvad-Udemy-100-days-of-code/048-the-coockie-factory/raw/branch/main/docs/images/step-4-buy-upgrade.png)

**Function**

```python
def choose_upgrade(items: Sequence[StoreItem]) -> StoreItem | None:
    affordable = [item for item in items if item.is_affordable]
    if not affordable:
        return None
    return max(affordable, key=lambda item: (item.price, item.index))
```

## What we gained

- **Keep the decision pure.** `choose_upgrade` takes plain `StoreItem` values and returns one, so it is tested with no game at all (ties, none affordable, empty list).
- **Confirm the result, do not assume it.** `buy_upgrade` checks that the owned amount (`productOwned<N>`) went up. Checking the cookie count first was wrong: the count keeps rising while the bot clicks.
- **The page lags behind.** The first release lost about 40 percent of its purchases: the game updates the price and owned text at once but the affordable flag only at its next redraw, so a second purchase in the same visit used a stale flag. Buying at most one product per store visit cut the failures to 1 of 49.
- **The real game finds what fakes cannot.** The stale-element retry and the owned-amount check both came from running in Docker.

Next: [Step 5: the five-minute run](Step-5-Five-Minute-Run)