Mogette · Chapter 5
Under the hood
Open data, one server, and the house.
The app is a small PHP server on my hosting and an installable web app in plain JavaScript. Everything it knows comes from open sources, my own accounts, or what I tell it.
Sources
| Source | What it gives | When |
|---|---|---|
| Lidl Plus | The offers of my shop, discounts on a selection, opening hours. | Every morning |
| Leaflet and lidl.fr | The week’s PDF, read offer by offer, and the in-store pages of the website. | Every morning |
| My Lidl Plus account | My coupons (read and activate), my receipts, my points. | Every morning, and on demand |
| FranceAgriMer | The average supermarket price of about 140 fresh products. | Every week |
| Open Prices | Shelf prices recorded by people, per chain, over 12 months. | When shown |
| Open Food Facts | The product behind a scanned barcode. | When scanning |
A scheduled task runs every morning at 5:00: it refreshes offers, coupons, receipts, points, opening hours and market prices. On shopping day the week is ready when I wake up. Lidl’s price history builds up each time the offers are read.
The app opens on the phone as well as on the computer. When two devices have changed the same week, it merges their changes, tick by tick for the list, instead of keeping only the last one saved.
Linked accounts
| Lidl Plus | I sign in myself, with my account. The server keeps the access to read my coupons and receipts and to activate the useful coupons. In the shop, the card shows inside the app. |
| Home Assistant | Through Mogi: the stock shared with Grocy, the phone notifications for bargains to freeze, and the one that starts sending recipes to the robot from the Mac. |
| Monsieur Cuisine | A script in my browser adds the recipes that have a robot version to my account, from my session. |
| Claude | The Mogette assistant, to write dishes around the week’s offers, with my API key. |
Through the method
The redesign went through the four phases of my method, with its agents. Every decision is logged with my words and the files it touched.
| Phase | What happened | Agent |
|---|---|---|
| Frame | My feedback and my two receipts, sorted into themes. The problem became “decide fast, and decide well”: say clearly what is worth buying, and say it when nothing can be compared. | Affinity map, Read the data, User journey, User story |
| Explore | Three visual directions on a canvas, five screens each. I kept the boldest one, then asked for the codes of discount shops: deep blue, yellow price tags, red prices. Copying Lidl’s logo and typeface was ruled out. | Ideation, Typography, Prototype |
| Build | The new app ran next to the old one until I switched. Checks before testing: 16 major accessibility issues and 3 blocking type issues, fixed first. | Components, Accessibility |
| Prove | On 6 November a heuristic review scored the app 21 out of 40, down from 26: features had piled up faster than the screens were rethought. Six tabs became three. Then four synthetic personas ran the main tasks on a phone; their results are labelled as synthetic. | Validate, Iteration, Deliver |
What it does not do
- It does not scrape other chains. Their sites are protected against bots, so the comparison relies on open data only, and most deals end with “no comparison possible”.
- It does not reuse Lidl’s logo or typeface: the look is that of discount shops, the name is its own.
- Prices outside the deals are estimates until a receipt confirms them; the app shows them with “≈”.
- It has one real user. The personas point to likely friction; only my own shopping trips prove anything.