Native? What native. Barcode scanning without a mobile app

There is a reflex, when someone says handheld warehouse app, to reach for React Native, two app stores, a release pipeline with signing keys, and a second codebase that drifts from the first. We have that reflex too — there is a half‑built native project in our repository to prove it. What the warehouse actually uses is the web application, on a tablet, with a scanner. Same URL as the office. This note is how that works and why it is enough.

The hardware

A mid‑range Android tablet in a rugged case on the trolley or the packing bench, and a Bluetooth barcode scanner paired as a keyboard. That last detail is the whole trick: a "keyboard wedge" scanner types the barcode, fast, and (usually) presses Enter. To the browser it is a person who types 13 digits in 40 milliseconds. No camera permissions, no SDK, no app.

Telling a scanner from a person

The picking and packing screens listen for keystrokes at the window level, not in a text box — a picker with gloves should not have to tap a field first. A small composable buffers keys and decides what it is looking at:

• Machine input arrives in a burst: the inter‑key gap is a few milliseconds. Human typing is tens to hundreds. Timing alone separates them.
• Enter terminates a scan on most scanners. Some are configured without it, so a quiet gap after a long enough burst is treated as the end of a code — long enough that it never submits half of a manual entry.
• Dialogs win. If a modal is open, the listener stands down; a scan into a confirmation dialog would be a bug.
• A queue, not a race. Scans are processed one at a time in order. A picker scanning four boxes in two seconds gets four results, in order, not three and a lost one.

When a code is recognised, the screen posts it — with the current picking session, the tab the picker is on, and optionally the location they are standing at — to one endpoint.

What the server does with a scan

One action, one round‑trip, and it answers a question that sounds simple and is not: what did you just scan, and what should happen?

1. Match. The code might be a product barcode, a trade‑unit barcode, an internal SKU, or a location label. Find the items in this session that could mean this code.
2. Choose. If several items match — the same product on two orders in the session — pick the one that is next in the route, unless the picker asked for a specific one.
3. Locate. Which location to pick from: the one the picker scanned, or the route's suggested one, or any that holds stock, in that order.
4. Pick. Record the pick in stock units, handling the case where the scanned barcode is a carton of six rather than a single unit.
5. Answer. A status (picked, already_picked, no_stock, not_found, wrong_state), a one‑line message in the picker's language — "Picked 6 × lavender candle from B‑11‑01 for 000071, 2 still to pick" — and just the row that changed, so the screen patches one line instead of reloading.

The same shape serves packing: scan an item at the bench, the server says which box it goes in and what is left.

Feeling like an app is a discipline

The tablet shows the same Vue application as a desk in the office; it feels like an app because of a handful of rules we hold to on warehouse screens and nowhere else:

• Full‑row targets. The whole row is the button. Nothing smaller than a thumb.
• Fewest taps to the common action, with the default already selected so one tap fires the label.
• Sound. A different tone for picked, already picked, not found. Nobody is looking at the screen.
• The screen never loses its place. A scan patches one row; the route and the tabs only change when the state changes.
• Pretty is allowed if it costs no taps. Morale is real; so is a queue of forty orders.

None of that needed a native framework. It needed people who stood at the bench for a day.

What we gave up, honestly

A web app cannot wake the device, cannot scan with the camera as smoothly as a native SDK (we do not try; the Bluetooth scanner is better anyway), and depends on Wi‑Fi coverage in the racks. We extended coverage. Offline picking is the one thing a native app would genuinely buy us, and so far the warehouse has never asked for it.

The payoff

One codebase, one deploy, one release number. A fix for the picking screen at ten in the morning is on every tablet at half past, with the rest of that day's releases. No store review, no "please update the app", no second team. When a new warehouse opens, the tablet setup is: open the browser, bookmark the URL, pair the scanner.