A short bar is not a bad month
Open a monthly chart on the 13th and the last bar is thirteen days tall next to thirty-day neighbours. It reads as a collapse. It is a calendar — and a chart that does not say so will have you ringing a supplier.
We built a chart this week that shows where the money came from over time — counter, QR ordering, delivery — stacked so you can see the mix change. Useful, ordinary, the sort of thing every system has.
It lies by default, and so does everybody else's.
The lie
Bucket by month. Open it today, the thirteenth. August is a bar built from thirty-one days of trade. September is a bar built from thirteen. September is less than half the height.
Nothing is broken. The numbers are right. But the picture says the month fell off a cliff, and the sentence a chart like that generates — "delivery down 57%" — is one somebody acts on. They ring the platform. They ask what happened to the listing. They change a price.
It happens at every zoom level, which is what makes it slippery. A weekly chart has the same problem every Sunday, when the current week is six days and a bit against a full seven. A daily chart has it all day, every day, because today is never over.
What a chart owes you
Three things, and none of them is difficult once you have decided to care.
Mark the unfinished period. Ours says Sep 2026 is still running in words under the chart, and the bar itself is drawn differently. Not a footnote — the thing a person reads first.
Leave it out of the comparison. When the chart works out whether delivery is up or down, it compares the last two complete periods. Today that means August against July. The month in progress is shown and not measured.
Say when there is nothing honest to say. A kitchen that opened eight days ago has no complete month. The chart says so, which is better than a confident percentage and much better than "no data" on a screen that plainly has some.
One we got wrong first
Our first version counted today as an elapsed day. So a week became "complete" one minute past midnight on its final Sunday, and the trend compared six days and a few hours against a full seven. The same lie, one zoom level down. Every test we had used months, so nothing touched it, and it was a screenshot that gave it away.
The rule that survived: a period is finished when its last day is behind us. Not when it has started. Not when most of it has gone.
What the chart will not tell you
Whether delivery growing is good news. That depends on a margin the chart has never been shown, and it is a decision about your business rather than a sentence a bar chart gets to write. Ours reports the move and stops.