Never installable

For three days of real-device testing, the app seemed to work, passing every test and review. However, it never once was installable on a real phone, and never indicated this. Built by a single technical person and an NHS dietetic team, with no patient data and no data leaving the device, it aimed to streamline manual keying into Nutritics, the nutrition-analysis software they already used.

The app relied on clients sending photos and descriptions of their meals, which the dietitian would then match to products and key into Nutritics.

The trial aimed to make the manual keying as fast and unambiguous as possible, serving as evidence to determine if Libro, the client-facing app, should be enabled. The app's limitations were discovered through real-world testing. The barcode scanning failed, as the crowdsourced database Open Food Facts could only name products and not provide nutrition information. The app drew a hard line at this point, reading barcodes but never writing nutrition figures.

The food library was designed to split distinct foods from daily logs, optimizing for the number of searches needed rather than the total number of logged eating occasions. A week's diary might only need thirty-five searches, as the same handful of items repeated across meals. This design choice reduced the workload on dietitians significantly, from a hundred and twenty lookups to just thirty-five.

However, the app faced numerous bugs when running on a real phone. The front end was a vanilla-JS progressive web app with no build toolchain, leading to issues when downloading data. IndexedDB round-trips caused delays, resulting in downloads that never appeared. Despite these challenges, the app demonstrated the potential to streamline manual data entry, paving the way for the integration of Libro in the future.