Capstone: A Complete, Working Food Tracker (Prisma Edition)

Marcus scans a carton of milk. Chapter 4's useBarcodeScanner hook decodes the barcode entirely client-side — genuinely unchanged, character for character, from both the Firebase and raw-SQL siblings — then handleDetected calls fetch("/api/lookup/..."). Chapter 3's route checks BarcodeCache first via findUnique() and falls back to a live Open Food Facts call, writing the result with upsert() — the fix for a real race condition the sibling's own plain INSERT doesn't have.

Chapter 5's AddItemForm pre-fills the name and category from that lookup. Marcus sets the expiry date and submits — the request body is { name, barcode, category, expiryDate }, the same field name end to end with no snake_case translation needed. Chapter 2's prisma.item.create() is the real validation gate, and it responds with the full created row — status and addedAt included, a real correctness detail the sibling's own hand-assembled response actually misses.

Chapter 6's alerts route surfaces the milk once it's within three days of its expiry date — expiryDate: { not: null, lte: threeDaysFromNow }, comparing a real typed DateTime rather than leaning on every date being formatted consistently by hand. The dashboard shows item.expiryDate.slice(0, 10) — the one small display-formatting cost of that same type safety.

Wanting to check if he's bought a particular brand of yogurt before, Marcus types "yog" into the search box. Chapter 7's useDebouncedSearch waits for him to stop typing, then hits prisma.item.findMany({ where: { name: { contains: q } } }) — matching case-insensitively only because Chapter 7's own migration added COLLATE NOCASE to the name column by hand. Without that fix, searching "yog" would never have matched an item actually named "Yogurt" — a real regression the sibling course never had to fix at all.

The milk gets finished. Chapter 8's PATCH /api/items/:id/use — never a plain link — calls prisma.item.update({ where: { id, status: "active" }, data: {...} }), the same atomic single-query guard as the sibling's own WHERE id = ? AND status = 'active', wrapped in try/catch to turn a thrown P2025 into a clean 404 if Marcus somehow clicks twice.

Marking the milk used calls Chapter 10's notifyChange() — one call, with no knowledge of who's listening, genuinely unchanged from the sibling course. The alerts dashboard disappears the milk from its own list; the recipe suggestions (Chapter 9) stop counting it as an expiring ingredient — both react on their own, entirely independently of whether the route they each call runs a raw SQL query or a Prisma one underneath.

With chicken and spinach both nearing expiry, Chapter 9's Promise.all fan-out queries TheMealDB for both concurrently — a real Node/JavaScript fact, identical in this course and its raw-SQL sibling, owing nothing to Prisma either way. RecipeCache reuses Chapter 3's own upsert() pattern for the same race-safety reason, storing each meal list as a plain JSON-encoded string rather than depending on Prisma's own inconsistent Json-type support for SQLite.

Marcus's session happens against a real deployment: Express serving the built React app via Chapter 11's own catch-all SPA route, pm2 keeping the process alive, nginx terminating HTTPS in front of it — all identical to the sibling. What isn't identical: npx prisma generate ran on the deploy server to rebuild the generated client, and npx prisma migrate deploy applied every migration this course's own eleven chapters accumulated, including Chapter 7's own hand-edited COLLATE NOCASE fix — with the underlying SQLite file's own persistence tracked through DATABASE_URL rather than a hardcoded path.