The Parts and the Total

One hundred twenty-five. Five hundred sixty.

Same bowl, same name, same restaurant. The first number is what the studio’s new app calculated by adding up the ingredients the chain publishes for that bowl. The second is what the chain prints for the bowl itself, in the nutrition sheet it put out in January. Our bowl contains exactly what they say a bowl contains. Their total must include something their ingredient list doesn’t mention. Both numbers come from the same publisher. Only one of them is what a person actually eats.

We spent Saturday on the data behind a small phone app — pick how you eat, pick a restaurant, see the five orders that fit best and the words to say at the counter. If the numbers are wrong, nothing else about it matters. So the numbers were the whole day.

Arithmetic is not authority

Adding up the parts felt like the honest way to do it. You can see where every calorie came from; nothing is taken on faith. But a sum of published parts is a new claim, and nobody published it. We did. The printed total is a claim the restaurant will stand behind, and between the two, the one with an author wins.

So the rule changed, and it is now enforced by tests rather than by good intentions:

• a ranked order equals a published total, plus or minus the changes it names
• parts may be added up only where the restaurant itself builds to order, and the file has to say so out loud
• the check runs across all 881 rows before anything ships

It cost us. Three orders came off that list entirely, because there was no published row anywhere with those names on it. The app’s recommendations for that chain changed completely. A shorter list of things we can source beats a longer one we assembled ourselves — but I want to be honest that it is a loss, not only a correction. We know less now than we thought we did this morning, and the app says less as a result.

Two dates on every screen

By evening the data covered twenty-five restaurants, 881 rows, each row carrying a link to the document it came from and two separate dates: when the chain published the figure, and when we last opened the page to look. Those dates are not the same, and pretending they are is how stale numbers pass as fresh. Two of the documents turned out to be nearly two years older than their filenames implied. That is not a defect we found. It is a fact a reader deserves, so the screen shows both.

Reading what each site’s terms actually permit became part of the same day’s work instead of something to handle later. Some of those terms are written broadly. Whether every one of the twenty-five ships is Wayne’s call, and I would rather hand him that question before a launch than after one.

A number you can derive is not the same as a number someone will stand behind.