“No overdue invoices” and “we couldn't check” look the same on most screens. We treat that as a bug.
Picture the office on a Monday morning. Someone opens the overdue invoices screen and it says there are none. Good news, so nobody chases anyone, and the week moves on. Now picture the same screen on a morning when the list failed to load: the connection dropped, a request timed out, something behind the page fell over. In a lot of software the two mornings look identical. A tidy blank space and a line that says “Nothing to show.”
One of those screens is telling the truth. The other is guessing, and the guess happens to be the most comforting answer available. This is a note on why we treat that as a bug in our own software, how we found we were doing it, and how you can test any tool you rely on for the same thing in about a minute.
An empty list is a statement
It is easy to think of an empty screen as the absence of information. It isn't. “No overdue invoices” is a claim about your business. So is “no jobs today,” “no new enquiries” and “no unassigned visits.” People act on those claims. They don't make the phone call, they don't check the roster, they tell the owner everything is in hand.
An error message gets questioned. A wrong number often gets noticed, because it disagrees with something the person already knows. An empty list gets believed, because it looks exactly like a quiet day, and everyone hopes for a quiet day. That makes it the most dangerous way for a screen to be wrong.
How it happens, and why it's so common
Nobody sets out to build this. It falls out of the shortest way to write a page. The code asks for a list, and if nothing comes back, it carries on with an empty one so the page doesn't crash. Then the page draws whatever it has. Zero rows from a failed request and zero rows from a real answer arrive at the screen looking the same, so they get drawn the same.
It also survives testing, for a plain reason. When a developer checks the page, the request works, because on a developer's desk it nearly always does. The failure only turns up in real conditions: a worker on one bar of signal in a basement car park, an office wifi that drops for ten seconds, a busy moment on the server. Those are the conditions your business runs in, and they are the ones a demo never shows you.
How we found it in our own software
We found pages in Dispatch doing exactly this. Not through a customer complaint, but during a hard review of the screens that deal with money, where we were reading our own code looking for ways it could mislead the office. A page that could not read its data was drawing the same friendly empty state as a page with nothing to show. Nothing had crashed, and no error had been logged anywhere a person would look. That is what makes this class of fault hard to see: from the outside, it looks like a page working.
The fix for one page is small. The page has to keep hold of whether the read worked, and draw a different thing when it didn't: a plain “we couldn't load this, try again” in place of the empty state. But we have written before that the fix is not the end of the bug. One corrected page does nothing about the next page written the short way. So the work that mattered was the check that came after: an automated one that runs with the rest of our tests and looks for pages that draw an empty state without first asking whether the read failed. It started with the screens staff use on their phones, because that is where patchy signal is normal, and it has been widening since.
The phone app had a wrinkle worth mentioning. Staff pages keep a saved copy so a worker can still see their day with no signal. A “couldn't load” screen must never be saved as that copy, or the worker would be shown yesterday's failure as today's roster. So the failed state is marked, and the part of the app that saves copies refuses anything carrying the mark. That took longer than the original fix. It usually does.
The three answers a screen can give
The rule we now hold a page to is that it must be able to say three different things, and never blur them.
- Here is what I found. The normal case.
- I looked, and there is nothing. A real empty state, which the office can act on with confidence.
- I couldn't look. Not an empty state. It says so plainly, and offers a way to try again.
There is a fourth, for anything that can work without a connection: this is what I had when I last could look, and here is when that was. A saved copy is useful. A saved copy passed off as live is the same lie with an older date on it.
The most useful thing a screen can say is sometimes “I don't know.” Software that can't say it will say something else instead.
Test your own tools in a minute
You don't need to read any code to find out how the software you rely on behaves. You need a phone and aeroplane mode.
- Open a screen that matters and that you know has something on it: unpaid invoices, today's jobs, new enquiries.
- Switch the phone to aeroplane mode, or turn off the wifi on the computer, and reload or open the screen fresh.
- Read what it says. “Couldn't load” or “you're offline” is the honest answer. Old information with a clear “last updated” time is fine too. A clean “nothing here” is the one to worry about.
- Try it on the screens your field staff use, because they are the people who meet bad signal every day.
If a tool fails this, it doesn't mean throw it out. It means you now know which of its quiet screens you can't take at face value, and that is worth knowing before a morning when it matters. Tell the vendor. It is a fair thing to ask them to fix, and how they respond will tell you something.
Where we land on it
It is the same standard we apply to AI, where we argued you should test what it does when it doesn't know. A person in your office who didn't know whether an invoice had been paid would say so. They wouldn't tell you nothing was owing because the filing cabinet was locked. Software should be held to that, and in Dispatch we hold ours to it, including the pages where we got it wrong first.
For this week: pick the one screen your business trusts most when it is empty, and run the aeroplane mode test on it. If it tells you the truth, good. If it tells you everything is fine, you have learned that its good news needs a second look.
Software for service businesses, built by an operator.
Job management, books, and AI agents that know your business.
Know people who run service businesses? Sell Dispatch with us and earn commission on every customer you bring in, for as long as they stay. Commission only, under your own ABN.